The OCaml system
release 5.1
Documentation and user’s manual
Xavier Leroy,
Damien Doligez, Alain Frisch, Jacques Garrigue,
Didier RĂ©my, KC Sivaramakrishnan and JĂ©rĂŽme Vouillon
8
thDecember , 2023
  Copyright © 2023 Institut National de Recherche en Informatique et en Automatique

This manual is also available in PDF, plain text, as a bundle of HTML files, and as a bundle of Emacs Info files.

Contents

Foreword

This manual documents the release 5.1 of the OCaml system. It is organized as follows.

Conventions

OCaml runs on several operating systems. The parts of this manual that are specific to one operating system are presented as shown below:

Unix:  This is material specific to the Unix family of operating systems, including Linux and macOS.
Windows:  This is material specific to Microsoft Windows (Vista, 7, 8, 10, 11).

License

The OCaml system is copyright © 1996–2023 Institut National de Recherche en Informatique et en Automatique (INRIA). INRIA holds all ownership rights to the OCaml system.

The OCaml system is open source and can be freely redistributed. See the file LICENSE in the distribution for licensing information.

The OCaml documentation and user’s manual is copyright © 2023 Institut National de Recherche en Informatique et en Automatique (INRIA).

The OCaml documentation and user's manual is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.

The sample code in the user's manual and in the reference documentation of the standard library is licensed under a Creative Commons CC0 1.0 Universal (CC0 1.0) Public Domain Dedication License.

Availability

The complete OCaml distribution can be accessed via the ocaml.org website. This site contains a lot of additional information on OCaml.

Part I
An introduction to OCaml

Chapter 1 The core language

This part of the manual is a tutorial introduction to the OCaml language. A good familiarity with programming in a conventional languages (say, C or Java) is assumed, but no prior exposure to functional languages is required. The present chapter introduces the core language. Chapter 2 deals with the module system, chapter 3 with the object-oriented features, chapter 4 with labeled arguments, chapter 5 with polymorphic variants, chapter 6 with the limitations of polymorphism, and chapter 8 gives some advanced examples.

1.1 Basics

For this overview of OCaml, we use the interactive system, which is started by running ocaml from the Unix shell or Windows command prompt. This tutorial is presented as the transcript of a session with the interactive system: lines starting with # represent user input; the system responses are printed below, without a leading #.

Under the interactive system, the user types OCaml phrases terminated by ;; in response to the # prompt, and the system compiles them on the fly, executes them, and prints the outcome of evaluation. Phrases are either simple expressions, or let definitions of identifiers (either values or functions).

# 1 + 2 * 3;;
- : int = 7
# let pi = 4.0 *. atan 1.0;;
val pi : float = 3.14159265358979312
# let square x = x *. x;;
val square : float -> float = <fun>
# square (sin pi) +. square (cos pi);;
- : float = 1.

The OCaml system computes both the value and the type for each phrase. Even function parameters need no explicit type declaration: the system infers their types from their usage in the function. Notice also that integers and floating-point numbers are distinct types, with distinct operators: + and * operate on integers, but +. and *. operate on floats.

# 1.0 * 2;;
Error: This expression has type float but an expression was expected of type int

Recursive functions are defined with the let rec binding:

# let rec fib n = if n < 2 then n else fib (n - 1) + fib (n - 2);;
val fib : int -> int = <fun>
# fib 10;;
- : int = 55

1.2 Data types

In addition to integers and floating-point numbers, OCaml offers the usual basic data types:

Predefined data structures include tuples, arrays, and lists. There are also general mechanisms for defining your own data structures, such as records and variants, which will be covered in more detail later; for now, we concentrate on lists. Lists are either given in extension as a bracketed list of semicolon-separated elements, or built from the empty list [] (pronounce “nil”) by adding elements in front using the :: (“cons”) operator.

# let l = ["is"; "a"; "tale"; "told"; "etc."];;
val l : string list = ["is"; "a"; "tale"; "told"; "etc."]
# "Life" :: l;;
- : string list = ["Life"; "is"; "a"; "tale"; "told"; "etc."]

As with all other OCaml data structures, lists do not need to be explicitly allocated and deallocated from memory: all memory management is entirely automatic in OCaml. Similarly, there is no explicit handling of pointers: the OCaml compiler silently introduces pointers where necessary.

As with most OCaml data structures, inspecting and destructuring lists is performed by pattern-matching. List patterns have exactly the same form as list expressions, with identifiers representing unspecified parts of the list. As an example, here is insertion sort on a list:

# let rec sort lst = match lst with [] -> [] | head :: tail -> insert head (sort tail) and insert elt lst = match lst with [] -> [elt] | head :: tail -> if elt <= head then elt :: lst else head :: insert elt tail ;;
val sort : 'a list -> 'a list = <fun> val insert : 'a -> 'a list -> 'a list = <fun>
# sort l;;
- : string list = ["a"; "etc."; "is"; "tale"; "told"]

The type inferred for sort, 'a list -> 'a list, means that sort can actually apply to lists of any type, and returns a list of the same type. The type 'a is a type variable, and stands for any given type. The reason why sort can apply to lists of any type is that the comparisons (=, <=, etc.) are polymorphic in OCaml: they operate between any two values of the same type. This makes sort itself polymorphic over all list types.

# sort [6; 2; 5; 3];;
- : int list = [2; 3; 5; 6]
# sort [3.14; 2.718];;
- : float list = [2.718; 3.14]

The sort function above does not modify its input list: it builds and returns a new list containing the same elements as the input list, in ascending order. There is actually no way in OCaml to modify a list in-place once it is built: we say that lists are immutable data structures. Most OCaml data structures are immutable, but a few (most notably arrays) are mutable, meaning that they can be modified in-place at any time.

The OCaml notation for the type of a function with multiple arguments is
arg1_type -> arg2_type -> ... -> return_type. For example, the type inferred for insert, 'a -> 'a list -> 'a list, means that insert takes two arguments, an element of any type 'a and a list with elements of the same type 'a and returns a list of the same type.

1.3 Functions as values

OCaml is a functional language: functions in the full mathematical sense are supported and can be passed around freely just as any other piece of data. For instance, here is a deriv function that takes any float function as argument and returns an approximation of its derivative function:

# let deriv f dx = fun x -> (f (x +. dx) -. f x) /. dx;;
val deriv : (float -> float) -> float -> float -> float = <fun>
# let sin' = deriv sin 1e-6;;
val sin' : float -> float = <fun>
# sin' pi;;
- : float = -1.00000000013961143

Even function composition is definable:

# let compose f g = fun x -> f (g x);;
val compose : ('a -> 'b) -> ('c -> 'a) -> 'c -> 'b = <fun>
# let cos2 = compose square cos;;
val cos2 : float -> float = <fun>

Functions that take other functions as arguments are called “functionals”, or “higher-order functions”. Functionals are especially useful to provide iterators or similar generic operations over a data structure. For instance, the standard OCaml library provides a List.map functional that applies a given function to each element of a list, and returns the list of the results:

# List.map (fun n -> n * 2 + 1) [0;1;2;3;4];;
- : int list = [1; 3; 5; 7; 9]

This functional, along with a number of other list and array functionals, is predefined because it is often useful, but there is nothing magic with it: it can easily be defined as follows.

# let rec map f l = match l with [] -> [] | hd :: tl -> f hd :: map f tl;;
val map : ('a -> 'b) -> 'a list -> 'b list = <fun>

1.4 Records and variants

User-defined data structures include records and variants. Both are defined with the type declaration. Here, we declare a record type to represent rational numbers.

# type ratio = {num: int; denom: int};;
type ratio = { num : int; denom : int; }
# let add_ratio r1 r2 = {num = r1.num * r2.denom + r2.num * r1.denom; denom = r1.denom * r2.denom};;
val add_ratio : ratio -> ratio -> ratio = <fun>
# add_ratio {num=1; denom=3} {num=2; denom=5};;
- : ratio = {num = 11; denom = 15}

Record fields can also be accessed through pattern-matching:

# let integer_part r = match r with {num=num; denom=denom} -> num / denom;;
val integer_part : ratio -> int = <fun>

Since there is only one case in this pattern matching, it is safe to expand directly the argument r in a record pattern:

# let integer_part {num=num; denom=denom} = num / denom;;
val integer_part : ratio -> int = <fun>

Unneeded fields can be omitted:

# let get_denom {denom=denom} = denom;;
val get_denom : ratio -> int = <fun>

Optionally, missing fields can be made explicit by ending the list of fields with a trailing wildcard _::

# let get_num {num=num; _ } = num;;
val get_num : ratio -> int = <fun>

When both sides of the = sign are the same, it is possible to avoid repeating the field name by eliding the =field part:

# let integer_part {num; denom} = num / denom;;
val integer_part : ratio -> int = <fun>

This short notation for fields also works when constructing records:

# let ratio num denom = {num; denom};;
val ratio : int -> int -> ratio = <fun>

At last, it is possible to update few fields of a record at once:

# let integer_product integer ratio = { ratio with num = integer * ratio.num };;
val integer_product : int -> ratio -> ratio = <fun>

With this functional update notation, the record on the left-hand side of with is copied except for the fields on the right-hand side which are updated.

The declaration of a variant type lists all possible forms for values of that type. Each case is identified by a name, called a constructor, which serves both for constructing values of the variant type and inspecting them by pattern-matching. Constructor names are capitalized to distinguish them from variable names (which must start with a lowercase letter). For instance, here is a variant type for doing mixed arithmetic (integers and floats):

# type number = Int of int | Float of float | Error;;
type number = Int of int | Float of float | Error

This declaration expresses that a value of type number is either an integer, a floating-point number, or the constant Error representing the result of an invalid operation (e.g. a division by zero).

Enumerated types are a special case of variant types, where all alternatives are constants:

# type sign = Positive | Negative;;
type sign = Positive | Negative
# let sign_int n = if n >= 0 then Positive else Negative;;
val sign_int : int -> sign = <fun>

To define arithmetic operations for the number type, we use pattern-matching on the two numbers involved:

# let add_num n1 n2 = match (n1, n2) with (Int i1, Int i2) -> (* Check for overflow of integer addition *) if sign_int i1 = sign_int i2 && sign_int (i1 + i2) <> sign_int i1 then Float(float i1 +. float i2) else Int(i1 + i2) | (Int i1, Float f2) -> Float(float i1 +. f2) | (Float f1, Int i2) -> Float(f1 +. float i2) | (Float f1, Float f2) -> Float(f1 +. f2) | (Error, _) -> Error | (_, Error) -> Error;;
val add_num : number -> number -> number = <fun>
# add_num (Int 123) (Float 3.14159);;
- : number = Float 126.14159

Another interesting example of variant type is the built-in 'a option type which represents either a value of type 'a or an absence of value:

# type 'a option = Some of 'a | None;;
type 'a option = Some of 'a | None

This type is particularly useful when defining function that can fail in common situations, for instance

# let safe_square_root x = if x >= 0. then Some(sqrt x) else None;;
val safe_square_root : float -> float option = <fun>

The most common usage of variant types is to describe recursive data structures. Consider for example the type of binary trees:

# type 'a btree = Empty | Node of 'a * 'a btree * 'a btree;;
type 'a btree = Empty | Node of 'a * 'a btree * 'a btree

This definition reads as follows: a binary tree containing values of type 'a (an arbitrary type) is either empty, or is a node containing one value of type 'a and two subtrees also containing values of type 'a, that is, two 'a btree.

Operations on binary trees are naturally expressed as recursive functions following the same structure as the type definition itself. For instance, here are functions performing lookup and insertion in ordered binary trees (elements increase from left to right):

# let rec member x btree = match btree with Empty -> false | Node(y, left, right) -> if x = y then true else if x < y then member x left else member x right;;
val member : 'a -> 'a btree -> bool = <fun>
# let rec insert x btree = match btree with Empty -> Node(x, Empty, Empty) | Node(y, left, right) -> if x <= y then Node(y, insert x left, right) else Node(y, left, insert x right);;
val insert : 'a -> 'a btree -> 'a btree = <fun>

1.4.1 Record and variant disambiguation

( This subsection can be skipped on the first reading )

Astute readers may have wondered what happens when two or more record fields or constructors share the same name

# type first_record = { x:int; y:int; z:int } type middle_record = { x:int; z:int } type last_record = { x:int };;
# type first_variant = A | B | C type last_variant = A;;

The answer is that when confronted with multiple options, OCaml tries to use locally available information to disambiguate between the various fields and constructors. First, if the type of the record or variant is known, OCaml can pick unambiguously the corresponding field or constructor. For instance:

# let look_at_x_then_z (r:first_record) = let x = r.x in x + r.z;;
val look_at_x_then_z : first_record -> int = <fun>
# let permute (x:first_variant) = match x with | A -> (B:first_variant) | B -> A | C -> C;;
val permute : first_variant -> first_variant = <fun>
# type wrapped = First of first_record let f (First r) = r, r.x;;
type wrapped = First of first_record val f : wrapped -> first_record * int = <fun>

In the first example, (r:first_record) is an explicit annotation telling OCaml that the type of r is first_record. With this annotation, Ocaml knows that r.x refers to the x field of the first record type. Similarly, the type annotation in the second example makes it clear to OCaml that the constructors A, B and C come from the first variant type. Contrarily, in the last example, OCaml has inferred by itself that the type of r can only be first_record and there are no needs for explicit type annotations.

Those explicit type annotations can in fact be used anywhere. Most of the time they are unnecessary, but they are useful to guide disambiguation, to debug unexpected type errors, or combined with some of the more advanced features of OCaml described in later chapters.

Secondly, for records, OCaml can also deduce the right record type by looking at the whole set of fields used in a expression or pattern:

# let project_and_rotate {x; y; _} = { x= - y; y = x; z = 0} ;;
val project_and_rotate : first_record -> first_record = <fun>

Since the fields x and y can only appear simultaneously in the first record type, OCaml infers that the type of project_and_rotate is first_record -> first_record.

In last resort, if there is not enough information to disambiguate between different fields or constructors, Ocaml picks the last defined type amongst all locally valid choices:

# let look_at_xz {x; z} = x;;
val look_at_xz : middle_record -> int = <fun>

Here, OCaml has inferred that the possible choices for the type of {x;z} are first_record and middle_record, since the type last_record has no field z. Ocaml then picks the type middle_record as the last defined type between the two possibilities.

Beware that this last resort disambiguation is local: once Ocaml has chosen a disambiguation, it sticks to this choice, even if it leads to an ulterior type error:

# let look_at_x_then_y r = let x = r.x in (* Ocaml deduces [r: last_record] *) x + r.y;;
Error: This expression has type last_record There is no field y within type last_record
# let is_a_or_b x = match x with | A -> true (* OCaml infers [x: last_variant] *) | B -> true;;
Error: This variant pattern is expected to have type last_variant There is no constructor B within type last_variant

Moreover, being the last defined type is a quite unstable position that may change surreptitiously after adding or moving around a type definition, or after opening a module (see chapter 2). Consequently, adding explicit type annotations to guide disambiguation is more robust than relying on the last defined type disambiguation.

1.5 Imperative features

Though all examples so far were written in purely applicative style, OCaml is also equipped with full imperative features. This includes the usual while and for loops, as well as mutable data structures such as arrays. Arrays are either created by listing semicolon-separated element values between [| and |] brackets, or allocated and initialized with the Array.make function, then filled up later by assignments. For instance, the function below sums two vectors (represented as float arrays) componentwise.

# let add_vect v1 v2 = let len = min (Array.length v1) (Array.length v2) in let res = Array.make len 0.0 in for i = 0 to len - 1 do res.(i) <- v1.(i) +. v2.(i) done; res;;
val add_vect : float array -> float array -> float array = <fun>
# add_vect [| 1.0; 2.0 |] [| 3.0; 4.0 |];;
- : float array = [|4.; 6.|]

Record fields can also be modified by assignment, provided they are declared mutable in the definition of the record type:

# type mutable_point = { mutable x: float; mutable y: float };;
type mutable_point = { mutable x : float; mutable y : float; }
# let translate p dx dy = p.x <- p.x +. dx; p.y <- p.y +. dy;;
val translate : mutable_point -> float -> float -> unit = <fun>
# let mypoint = { x = 0.0; y = 0.0 };;
val mypoint : mutable_point = {x = 0.; y = 0.}
# translate mypoint 1.0 2.0;;
- : unit = ()
# mypoint;;
- : mutable_point = {x = 1.; y = 2.}

OCaml has no built-in notion of variable – identifiers whose current value can be changed by assignment. (The let binding is not an assignment, it introduces a new identifier with a new scope.) However, the standard library provides references, which are mutable indirection cells, with operators ! to fetch the current contents of the reference and := to assign the contents. Variables can then be emulated by let-binding a reference. For instance, here is an in-place insertion sort over arrays:

# let insertion_sort a = for i = 1 to Array.length a - 1 do let val_i = a.(i) in let j = ref i in while !j > 0 && val_i < a.(!j - 1) do a.(!j) <- a.(!j - 1); j := !j - 1 done; a.(!j) <- val_i done;;
val insertion_sort : 'a array -> unit = <fun>

References are also useful to write functions that maintain a current state between two calls to the function. For instance, the following pseudo-random number generator keeps the last returned number in a reference:

# let current_rand = ref 0;;
val current_rand : int ref = {contents = 0}
# let random () = current_rand := !current_rand * 25713 + 1345; !current_rand;;
val random : unit -> int = <fun>

Again, there is nothing magical with references: they are implemented as a single-field mutable record, as follows.

# type 'a ref = { mutable contents: 'a };;
type 'a ref = { mutable contents : 'a; }
# let ( ! ) r = r.contents;;
val ( ! ) : 'a ref -> 'a = <fun>
# let ( := ) r newval = r.contents <- newval;;
val ( := ) : 'a ref -> 'a -> unit = <fun>

In some special cases, you may need to store a polymorphic function in a data structure, keeping its polymorphism. Doing this requires user-provided type annotations, since polymorphism is only introduced automatically for global definitions. However, you can explicitly give polymorphic types to record fields.

# type idref = { mutable id: 'a. 'a -> 'a };;
type idref = { mutable id : 'a. 'a -> 'a; }
# let r = {id = fun x -> x};;
val r : idref = {id = <fun>}
# let g s = (s.id 1, s.id true);;
val g : idref -> int * bool = <fun>
# r.id <- (fun x -> print_string "called id\n"; x);;
- : unit = ()
# g r;;
called id called id - : int * bool = (1, true)

1.6 Exceptions

OCaml provides exceptions for signalling and handling exceptional conditions. Exceptions can also be used as a general-purpose non-local control structure, although this should not be overused since it can make the code harder to understand. Exceptions are declared with the exception construct, and signalled with the raise operator. For instance, the function below for taking the head of a list uses an exception to signal the case where an empty list is given.

# exception Empty_list;;
exception Empty_list
# let head l = match l with [] -> raise Empty_list | hd :: tl -> hd;;
val head : 'a list -> 'a = <fun>
# head [1; 2];;
- : int = 1
# head [];;
Exception: Empty_list.

Exceptions are used throughout the standard library to signal cases where the library functions cannot complete normally. For instance, the List.assoc function, which returns the data associated with a given key in a list of (key, data) pairs, raises the predefined exception Not_found when the key does not appear in the list:

# List.assoc 1 [(0, "zero"); (1, "one")];;
- : string = "one"
# List.assoc 2 [(0, "zero"); (1, "one")];;
Exception: Not_found.

Exceptions can be trapped with the try
with construct:

# let name_of_binary_digit digit = try List.assoc digit [0, "zero"; 1, "one"] with Not_found -> "not a binary digit";;
val name_of_binary_digit : int -> string = <fun>
# name_of_binary_digit 0;;
- : string = "zero"
# name_of_binary_digit (-1);;
- : string = "not a binary digit"

The with part does pattern matching on the exception value with the same syntax and behavior as match. Thus, several exceptions can be caught by one try
with construct:

# let rec first_named_value values names = try List.assoc (head values) names with | Empty_list -> "no named value" | Not_found -> first_named_value (List.tl values) names;;
val first_named_value : 'a list -> ('a * string) list -> string = <fun>
# first_named_value [0; 10] [1, "one"; 10, "ten"];;
- : string = "ten"

Also, finalization can be performed by trapping all exceptions, performing the finalization, then re-raising the exception:

# let temporarily_set_reference ref newval funct = let oldval = !ref in try ref := newval; let res = funct () in ref := oldval; res with x -> ref := oldval; raise x;;
val temporarily_set_reference : 'a ref -> 'a -> (unit -> 'b) -> 'b = <fun>

An alternative to try
with is to catch the exception while pattern matching:

# let assoc_may_map f x l = match List.assoc x l with | exception Not_found -> None | y -> f y;;
val assoc_may_map : ('a -> 'b option) -> 'c -> ('c * 'a) list -> 'b option = <fun>

Note that this construction is only useful if the exception is raised between match
with. Exception patterns can be combined with ordinary patterns at the toplevel,

# let flat_assoc_opt x l = match List.assoc x l with | None | exception Not_found -> None | Some _ as v -> v;;
val flat_assoc_opt : 'a -> ('a * 'b option) list -> 'b option = <fun>

but they cannot be nested inside other patterns. For instance, the pattern Some (exception A) is invalid.

When exceptions are used as a control structure, it can be useful to make them as local as possible by using a locally defined exception. For instance, with

# let fixpoint f x = let exception Done in let x = ref x in try while true do let y = f !x in if !x = y then raise Done else x := y done; assert false with Done -> !x;;
val fixpoint : ('a -> 'a) -> 'a -> 'a = <fun>

the function f cannot raise a Done exception, which removes an entire class of misbehaving functions.

1.7 Lazy expressions

OCaml allows us to defer some computation until later when we need the result of that computation.

We use lazy (expr) to delay the evaluation of some expression expr. For example, we can defer the computation of 1+1 until we need the result of that expression, 2. Let us see how we initialize a lazy expression.

# let lazy_two = lazy (print_endline "lazy_two evaluation"; 1 + 1);;
val lazy_two : int lazy_t = <lazy>

We added print_endline "lazy_two evaluation" to see when the lazy expression is being evaluated.

The value of lazy_two is displayed as <lazy>, which means the expression has not been evaluated yet, and its final value is unknown.

Note that lazy_two has type int lazy_t. However, the type 'a lazy_t is an internal type name, so the type 'a Lazy.t should be preferred when possible.

When we finally need the result of a lazy expression, we can call Lazy.force on that expression to force its evaluation. The function force comes from standard-library module Lazy.

# Lazy.force lazy_two;;
lazy_two evaluation - : int = 2

Notice that our function call above prints “lazy_two evaluation” and then returns the plain value of the computation.

Now if we look at the value of lazy_two, we see that it is not displayed as <lazy> anymore but as lazy 2.

# lazy_two;;
- : int lazy_t = lazy 2

This is because Lazy.force memoizes the result of the forced expression. In other words, every subsequent call of Lazy.force on that expression returns the result of the first computation without recomputing the lazy expression. Let us force lazy_two once again.

# Lazy.force lazy_two;;
- : int = 2

The expression is not evaluated this time; notice that “lazy_two evaluation” is not printed. The result of the initial computation is simply returned.

Lazy patterns provide another way to force a lazy expression.

# let lazy_l = lazy ([1; 2] @ [3; 4]);;
val lazy_l : int list lazy_t = <lazy>
# let lazy l = lazy_l;;
val l : int list = [1; 2; 3; 4]

We can also use lazy patterns in pattern matching.

# let maybe_eval lazy_guard lazy_expr = match lazy_guard, lazy_expr with | lazy false, _ -> "matches if (Lazy.force lazy_guard = false); lazy_expr not forced" | lazy true, lazy _ -> "matches if (Lazy.force lazy_guard = true); lazy_expr forced";;
val maybe_eval : bool lazy_t -> 'a lazy_t -> string = <fun>

The lazy expression lazy_expr is forced only if the lazy_guard value yields true once computed. Indeed, a simple wildcard pattern (not lazy) never forces the lazy expression’s evaluation. However, a pattern with keyword lazy, even if it is wildcard, always forces the evaluation of the deferred computation.

1.8 Symbolic processing of expressions

We finish this introduction with a more complete example representative of the use of OCaml for symbolic processing: formal manipulations of arithmetic expressions containing variables. The following variant type describes the expressions we shall manipulate:

# type expression = Const of float | Var of string | Sum of expression * expression (* e1 + e2 *) | Diff of expression * expression (* e1 - e2 *) | Prod of expression * expression (* e1 * e2 *) | Quot of expression * expression (* e1 / e2 *) ;;
type expression = Const of float | Var of string | Sum of expression * expression | Diff of expression * expression | Prod of expression * expression | Quot of expression * expression

We first define a function to evaluate an expression given an environment that maps variable names to their values. For simplicity, the environment is represented as an association list.

# exception Unbound_variable of string;;
exception Unbound_variable of string
# let rec eval env exp = match exp with Const c -> c | Var v -> (try List.assoc v env with Not_found -> raise (Unbound_variable v)) | Sum(f, g) -> eval env f +. eval env g | Diff(f, g) -> eval env f -. eval env g | Prod(f, g) -> eval env f *. eval env g | Quot(f, g) -> eval env f /. eval env g;;
val eval : (string * float) list -> expression -> float = <fun>
# eval [("x", 1.0); ("y", 3.14)] (Prod(Sum(Var "x", Const 2.0), Var "y"));;
- : float = 9.42

Now for a real symbolic processing, we define the derivative of an expression with respect to a variable dv:

# let rec deriv exp dv = match exp with Const c -> Const 0.0 | Var v -> if v = dv then Const 1.0 else Const 0.0 | Sum(f, g) -> Sum(deriv f dv, deriv g dv) | Diff(f, g) -> Diff(deriv f dv, deriv g dv) | Prod(f, g) -> Sum(Prod(f, deriv g dv), Prod(deriv f dv, g)) | Quot(f, g) -> Quot(Diff(Prod(deriv f dv, g), Prod(f, deriv g dv)), Prod(g, g)) ;;
val deriv : expression -> string -> expression = <fun>
# deriv (Quot(Const 1.0, Var "x")) "x";;
- : expression = Quot (Diff (Prod (Const 0., Var "x"), Prod (Const 1., Const 1.)), Prod (Var "x", Var "x"))

1.9 Pretty-printing

As shown in the examples above, the internal representation (also called abstract syntax) of expressions quickly becomes hard to read and write as the expressions get larger. We need a printer and a parser to go back and forth between the abstract syntax and the concrete syntax, which in the case of expressions is the familiar algebraic notation (e.g. 2*x+1).

For the printing function, we take into account the usual precedence rules (i.e. * binds tighter than +) to avoid printing unnecessary parentheses. To this end, we maintain the current operator precedence and print parentheses around an operator only if its precedence is less than the current precedence.

# let print_expr exp = (* Local function definitions *) let open_paren prec op_prec = if prec > op_prec then print_string "(" in let close_paren prec op_prec = if prec > op_prec then print_string ")" in let rec print prec exp = (* prec is the current precedence *) match exp with Const c -> print_float c | Var v -> print_string v | Sum(f, g) -> open_paren prec 0; print 0 f; print_string " + "; print 0 g; close_paren prec 0 | Diff(f, g) -> open_paren prec 0; print 0 f; print_string " - "; print 1 g; close_paren prec 0 | Prod(f, g) -> open_paren prec 2; print 2 f; print_string " * "; print 2 g; close_paren prec 2 | Quot(f, g) -> open_paren prec 2; print 2 f; print_string " / "; print 3 g; close_paren prec 2 in print 0 exp;;
val print_expr : expression -> unit = <fun>
# let e = Sum(Prod(Const 2.0, Var "x"), Const 1.0);;
val e : expression = Sum (Prod (Const 2., Var "x"), Const 1.)
# print_expr e; print_newline ();;
2. * x + 1. - : unit = ()
# print_expr (deriv e "x"); print_newline ();;
2. * 1. + 0. * x + 0. - : unit = ()

1.10 Printf formats

There is a printf function in the Printf module (see chapter 2) that allows you to make formatted output more concisely. It follows the behavior of the printf function from the C standard library. The printf function takes a format string that describes the desired output as a text interspersed with specifiers (for instance %d, %f). Next, the specifiers are substituted by the following arguments in their order of apparition in the format string:

# Printf.printf "%i + %i is an integer value, %F * %F is a float, %S\n" 3 2 4.5 1. "this is a string";;
3 + 2 is an integer value, 4.5 * 1. is a float, "this is a string" - : unit = ()

The OCaml type system checks that the type of the arguments and the specifiers are compatible. If you pass it an argument of a type that does not correspond to the format specifier, the compiler will display an error message:

# Printf.printf "Float value: %F" 42;;
Error: This expression has type int but an expression was expected of type float Hint: Did you mean `42.'?

The fprintf function is like printf except that it takes an output channel as the first argument. The %a specifier can be useful to define custom printers (for custom types). For instance, we can create a printing template that converts an integer argument to signed decimal:

# let pp_int ppf n = Printf.fprintf ppf "%d" n;;
val pp_int : out_channel -> int -> unit = <fun>
# Printf.printf "Outputting an integer using a custom printer: %a " pp_int 42;;
Outputting an integer using a custom printer: 42 - : unit = ()

The advantage of those printers based on the %a specifier is that they can be composed together to create more complex printers step by step. We can define a combinator that can turn a printer for 'a type into a printer for 'a optional:

# let pp_option printer ppf = function | None -> Printf.fprintf ppf "None" | Some v -> Printf.fprintf ppf "Some(%a)" printer v;;
val pp_option : (out_channel -> 'a -> unit) -> out_channel -> 'a option -> unit = <fun>
# Printf.fprintf stdout "The current setting is %a. \nThere is only %a\n" (pp_option pp_int) (Some 3) (pp_option pp_int) None ;;
The current setting is Some(3). There is only None - : unit = ()

If the value of its argument its None, the printer returned by pp_option printer prints None otherwise it uses the provided printer to print Some .

Here is how to rewrite the pretty-printer using fprintf:

# let pp_expr ppf expr = let open_paren prec op_prec output = if prec > op_prec then Printf.fprintf output "%s" "(" in let close_paren prec op_prec output = if prec > op_prec then Printf.fprintf output "%s" ")" in let rec print prec ppf expr = match expr with | Const c -> Printf.fprintf ppf "%F" c | Var v -> Printf.fprintf ppf "%s" v | Sum(f, g) -> open_paren prec 0 ppf; Printf.fprintf ppf "%a + %a" (print 0) f (print 0) g; close_paren prec 0 ppf | Diff(f, g) -> open_paren prec 0 ppf; Printf.fprintf ppf "%a - %a" (print 0) f (print 1) g; close_paren prec 0 ppf | Prod(f, g) -> open_paren prec 2 ppf; Printf.fprintf ppf "%a * %a" (print 2) f (print 2) g; close_paren prec 2 ppf | Quot(f, g) -> open_paren prec 2 ppf; Printf.fprintf ppf "%a / %a" (print 2) f (print 3) g; close_paren prec 2 ppf in print 0 ppf expr;;
val pp_expr : out_channel -> expression -> unit = <fun>
# pp_expr stdout e; print_newline ();;
2. * x + 1. - : unit = ()
# pp_expr stdout (deriv e "x"); print_newline ();;
2. * 1. + 0. * x + 0. - : unit = ()

Due to the way that format strings are built, storing a format string requires an explicit type annotation:

# let str : _ format = "%i is an integer value, %F is a float, %S\n";;
# Printf.printf str 3 4.5 "string value";;
3 is an integer value, 4.5 is a float, "string value" - : unit = ()

1.11 Standalone OCaml programs

All examples given so far were executed under the interactive system. OCaml code can also be compiled separately and executed non-interactively using the batch compilers ocamlc and ocamlopt. The source code must be put in a file with extension .ml. It consists of a sequence of phrases, which will be evaluated at runtime in their order of appearance in the source file. Unlike in interactive mode, types and values are not printed automatically; the program must call printing functions explicitly to produce some output. The ;; used in the interactive examples is not required in source files created for use with OCaml compilers, but can be helpful to mark the end of a top-level expression unambiguously even when there are syntax errors. Here is a sample standalone program to print the greatest common divisor (gcd) of two numbers:

(* File gcd.ml *)
let rec gcd a b =
  if b = 0 then a
  else gcd b (a mod b);;

let main () =
  let a = int_of_string Sys.argv.(1) in
  let b = int_of_string Sys.argv.(2) in
  Printf.printf "%d\n" (gcd a b);
  exit 0;;
main ();;

Sys.argv is an array of strings containing the command-line parameters. Sys.argv.(1) is thus the first command-line parameter. The program above is compiled and executed with the following shell commands:

$ ocamlc -o gcd gcd.ml
$ ./gcd 6 9
3
$ ./gcd 7 11
1

More complex standalone OCaml programs are typically composed of multiple source files, and can link with precompiled libraries. Chapters 13 and 16 explain how to use the batch compilers ocamlc and ocamlopt. Recompilation of multi-file OCaml projects can be automated using third-party build systems, such as dune.

Chapter 2 The module system

This chapter introduces the module system of OCaml.

2.1 Structures

A primary motivation for modules is to package together related definitions (such as the definitions of a data type and associated operations over that type) and enforce a consistent naming scheme for these definitions. This avoids running out of names or accidentally confusing names. Such a package is called a structure and is introduced by the struct
end construct, which contains an arbitrary sequence of definitions. The structure is usually given a name with the module binding. For instance, here is a structure packaging together a type of FIFO queues and their operations:

# module Fifo = struct type 'a queue = { front: 'a list; rear: 'a list } let make front rear = match front with | [] -> { front = List.rev rear; rear = [] } | _ -> { front; rear } let empty = { front = []; rear = [] } let is_empty = function { front = []; _ } -> true | _ -> false let add x q = make q.front (x :: q.rear) exception Empty let top = function | { front = []; _ } -> raise Empty | { front = x :: _; _ } -> x let pop = function | { front = []; _ } -> raise Empty | { front = _ :: f; rear = r } -> make f r end;;
module Fifo : sig type 'a queue = { front : 'a list; rear : 'a list; } val make : 'a list -> 'a list -> 'a queue val empty : 'a queue val is_empty : 'a queue -> bool val add : 'a -> 'a queue -> 'a queue exception Empty val top : 'a queue -> 'a val pop : 'a queue -> 'a queue end

Outside the structure, its components can be referred to using the “dot notation”, that is, identifiers qualified by a structure name. For instance, Fifo.add is the function add defined inside the structure Fifo and Fifo.queue is the type queue defined in Fifo.

# Fifo.add "hello" Fifo.empty;;
- : string Fifo.queue = {Fifo.front = ["hello"]; rear = []}

Another possibility is to open the module, which brings all identifiers defined inside the module into the scope of the current structure.

# open Fifo;;
# add "hello" empty;;
- : string Fifo.queue = {front = ["hello"]; rear = []}

Opening a module enables lighter access to its components, at the cost of making it harder to identify in which module an identifier has been defined. In particular, opened modules can shadow identifiers present in the current scope, potentially leading to confusing errors:

# let empty = [] open Fifo;;
val empty : 'a list = []
# let x = 1 :: empty ;;
Error: This expression has type 'a Fifo.queue but an expression was expected of type int list

A partial solution to this conundrum is to open modules locally, making the components of the module available only in the concerned expression. This can also make the code both easier to read (since the open statement is closer to where it is used) and easier to refactor (since the code fragment is more self-contained). Two constructions are available for this purpose:

# let open Fifo in add "hello" empty;;
- : string Fifo.queue = {front = ["hello"]; rear = []}

and

# Fifo.(add "hello" empty);;
- : string Fifo.queue = {front = ["hello"]; rear = []}

In the second form, when the body of a local open is itself delimited by parentheses, braces or bracket, the parentheses of the local open can be omitted. For instance,

# Fifo.[empty] = Fifo.([empty]);;
- : bool = true
# Fifo.[|empty|] = Fifo.([|empty|]);;
- : bool = true
# Fifo.{ contents = empty } = Fifo.({ contents = empty });;
- : bool = true

This second form also works for patterns:

# let at_most_one_element x = match x with | Fifo.{ front = ([] | [_]); rear = [] } -> true | _ -> false ;;
val at_most_one_element : 'a Fifo.queue -> bool = <fun>

It is also possible to copy the components of a module inside another module by using an include statement. This can be particularly useful to extend existing modules. As an illustration, we could add functions that return an optional value rather than an exception when the queue is empty.

# module FifoOpt = struct include Fifo let top_opt q = if is_empty q then None else Some(top q) let pop_opt q = if is_empty q then None else Some(pop q) end;;
module FifoOpt : sig type 'a queue = 'a Fifo.queue = { front : 'a list; rear : 'a list; } val make : 'a list -> 'a list -> 'a queue val empty : 'a queue val is_empty : 'a queue -> bool val add : 'a -> 'a queue -> 'a queue exception Empty val top : 'a queue -> 'a val pop : 'a queue -> 'a queue val top_opt : 'a queue -> 'a option val pop_opt : 'a queue -> 'a queue option end

2.2 Signatures

Signatures are interfaces for structures. A signature specifies which components of a structure are accessible from the outside, and with which type. It can be used to hide some components of a structure (e.g. local function definitions) or export some components with a restricted type. For instance, the signature below specifies the queue operations empty, add, top and pop, but not the auxiliary function make. Similarly, it makes the queue type abstract (by not providing its actual representation as a concrete type). This ensures that users of the Fifo module cannot violate data structure invariants that operations rely on, such as “if the front list is empty, the rear list must also be empty”.

# module type FIFO = sig type 'a queue (* now an abstract type *) val empty : 'a queue val add : 'a -> 'a queue -> 'a queue val top : 'a queue -> 'a val pop : 'a queue -> 'a queue exception Empty end;;
module type FIFO = sig type 'a queue val empty : 'a queue val add : 'a -> 'a queue -> 'a queue val top : 'a queue -> 'a val pop : 'a queue -> 'a queue exception Empty end

Restricting the Fifo structure to this signature results in another view of the Fifo structure where the make function is not accessible and the actual representation of queues is hidden:

# module AbstractQueue = (Fifo : FIFO);;
module AbstractQueue : FIFO
# AbstractQueue.make [1] [2;3] ;;
Error: Unbound value AbstractQueue.make
# AbstractQueue.add "hello" AbstractQueue.empty;;
- : string AbstractQueue.queue = <abstr>

The restriction can also be performed during the definition of the structure, as in

module Fifo = (struct ... end : FIFO);;

An alternate syntax is provided for the above:

module Fifo : FIFO = struct ... end;;

Like for modules, it is possible to include a signature to copy its components inside the current signature. For instance, we can extend the FIFO signature with the top_opt and pop_opt functions:

# module type FIFO_WITH_OPT = sig include FIFO val top_opt: 'a queue -> 'a option val pop_opt: 'a queue -> 'a queue option end;;
module type FIFO_WITH_OPT = sig type 'a queue val empty : 'a queue val add : 'a -> 'a queue -> 'a queue val top : 'a queue -> 'a val pop : 'a queue -> 'a queue exception Empty val top_opt : 'a queue -> 'a option val pop_opt : 'a queue -> 'a queue option end

2.3 Functors

Functors are “functions” from modules to modules. Functors let you create parameterized modules and then provide other modules as parameter(s) to get a specific implementation. For instance, a Set module implementing sets as sorted lists could be parameterized to work with any module that provides an element type and a comparison function compare (such as OrderedString):

# type comparison = Less | Equal | Greater;;
type comparison = Less | Equal | Greater
# module type ORDERED_TYPE = sig type t val compare: t -> t -> comparison end;;
module type ORDERED_TYPE = sig type t val compare : t -> t -> comparison end
# module Set = functor (Elt: ORDERED_TYPE) -> struct type element = Elt.t type set = element list let empty = [] let rec add x s = match s with [] -> [x] | hd::tl -> match Elt.compare x hd with Equal -> s (* x is already in s *) | Less -> x :: s (* x is smaller than all elements of s *) | Greater -> hd :: add x tl let rec member x s = match s with [] -> false | hd::tl -> match Elt.compare x hd with Equal -> true (* x belongs to s *) | Less -> false (* x is smaller than all elements of s *) | Greater -> member x tl end;;
module Set : functor (Elt : ORDERED_TYPE) -> sig type element = Elt.t type set = element list val empty : 'a list val add : Elt.t -> Elt.t list -> Elt.t list val member : Elt.t -> Elt.t list -> bool end

By applying the Set functor to a structure implementing an ordered type, we obtain set operations for this type:

# module OrderedString = struct type t = string let compare x y = if x = y then Equal else if x < y then Less else Greater end;;
module OrderedString : sig type t = string val compare : 'a -> 'a -> comparison end
# module StringSet = Set(OrderedString);;
module StringSet : sig type element = OrderedString.t type set = element list val empty : 'a list val add : OrderedString.t -> OrderedString.t list -> OrderedString.t list val member : OrderedString.t -> OrderedString.t list -> bool end
# StringSet.member "bar" (StringSet.add "foo" StringSet.empty);;
- : bool = false

2.4 Functors and type abstraction

As in the Fifo example, it would be good style to hide the actual implementation of the type set, so that users of the structure will not rely on sets being lists, and we can switch later to another, more efficient representation of sets without breaking their code. This can be achieved by restricting Set by a suitable functor signature:

# module type SETFUNCTOR = functor (Elt: ORDERED_TYPE) -> sig type element = Elt.t (* concrete *) type set (* abstract *) val empty : set val add : element -> set -> set val member : element -> set -> bool end;;
module type SETFUNCTOR = functor (Elt : ORDERED_TYPE) -> sig type element = Elt.t type set val empty : set val add : element -> set -> set val member : element -> set -> bool end
# module AbstractSet = (Set : SETFUNCTOR);;
module AbstractSet : SETFUNCTOR
# module AbstractStringSet = AbstractSet(OrderedString);;
module AbstractStringSet : sig type element = OrderedString.t type set = AbstractSet(OrderedString).set val empty : set val add : element -> set -> set val member : element -> set -> bool end
# AbstractStringSet.add "gee" AbstractStringSet.empty;;
- : AbstractStringSet.set = <abstr>

In an attempt to write the type constraint above more elegantly, one may wish to name the signature of the structure returned by the functor, then use that signature in the constraint:

# module type SET = sig type element type set val empty : set val add : element -> set -> set val member : element -> set -> bool end;;
module type SET = sig type element type set val empty : set val add : element -> set -> set val member : element -> set -> bool end
# module WrongSet = (Set : functor(Elt: ORDERED_TYPE) -> SET);;
module WrongSet : functor (Elt : ORDERED_TYPE) -> SET
# module WrongStringSet = WrongSet(OrderedString);;
module WrongStringSet : sig type element = WrongSet(OrderedString).element type set = WrongSet(OrderedString).set val empty : set val add : element -> set -> set val member : element -> set -> bool end
# WrongStringSet.add "gee" WrongStringSet.empty ;;
Error: This expression has type string but an expression was expected of type WrongStringSet.element = WrongSet(OrderedString).element

The problem here is that SET specifies the type element abstractly, so that the type equality between element in the result of the functor and t in its argument is forgotten. Consequently, WrongStringSet.element is not the same type as string, and the operations of WrongStringSet cannot be applied to strings. As demonstrated above, it is important that the type element in the signature SET be declared equal to Elt.t; unfortunately, this is impossible above since SET is defined in a context where Elt does not exist. To overcome this difficulty, OCaml provides a with type construct over signatures that allows enriching a signature with extra type equalities:

# module AbstractSet2 = (Set : functor(Elt: ORDERED_TYPE) -> (SET with type element = Elt.t));;
module AbstractSet2 : functor (Elt : ORDERED_TYPE) -> sig type element = Elt.t type set val empty : set val add : element -> set -> set val member : element -> set -> bool end

As in the case of simple structures, an alternate syntax is provided for defining functors and restricting their result:

module AbstractSet2(Elt: ORDERED_TYPE) : (SET with type element = Elt.t) =
  struct ... end;;

Abstracting a type component in a functor result is a powerful technique that provides a high degree of type safety, as we now illustrate. Consider an ordering over character strings that is different from the standard ordering implemented in the OrderedString structure. For instance, we compare strings without distinguishing upper and lower case.

# module NoCaseString = struct type t = string let compare s1 s2 = OrderedString.compare (String.lowercase_ascii s1) (String.lowercase_ascii s2) end;;
module NoCaseString : sig type t = string val compare : string -> string -> comparison end
# module NoCaseStringSet = AbstractSet(NoCaseString);;
module NoCaseStringSet : sig type element = NoCaseString.t type set = AbstractSet(NoCaseString).set val empty : set val add : element -> set -> set val member : element -> set -> bool end
# NoCaseStringSet.add "FOO" AbstractStringSet.empty ;;
Error: This expression has type AbstractStringSet.set = AbstractSet(OrderedString).set but an expression was expected of type NoCaseStringSet.set = AbstractSet(NoCaseString).set

Note that the two types AbstractStringSet.set and NoCaseStringSet.set are not compatible, and values of these two types do not match. This is the correct behavior: even though both set types contain elements of the same type (strings), they are built upon different orderings of that type, and different invariants need to be maintained by the operations (being strictly increasing for the standard ordering and for the case-insensitive ordering). Applying operations from AbstractStringSet to values of type NoCaseStringSet.set could give incorrect results, or build lists that violate the invariants of NoCaseStringSet.

2.5 Modules and separate compilation

All examples of modules so far have been given in the context of the interactive system. However, modules are most useful for large, batch-compiled programs. For these programs, it is a practical necessity to split the source into several files, called compilation units, that can be compiled separately, thus minimizing recompilation after changes.

In OCaml, compilation units are special cases of structures and signatures, and the relationship between the units can be explained easily in terms of the module system. A compilation unit A comprises two files:

These two files together define a structure named A as if the following definition was entered at top-level:

module A: sig (* contents of file A.mli *) end
        = struct (* contents of file A.ml *) end;;

The files that define the compilation units can be compiled separately using the ocamlc -c command (the -c option means “compile only, do not try to link”); this produces compiled interface files (with extension .cmi) and compiled object code files (with extension .cmo). When all units have been compiled, their .cmo files are linked together using the ocamlc command. For instance, the following commands compile and link a program composed of two compilation units Aux and Main:

$ ocamlc -c Aux.mli                     # produces aux.cmi
$ ocamlc -c Aux.ml                      # produces aux.cmo
$ ocamlc -c Main.mli                    # produces main.cmi
$ ocamlc -c Main.ml                     # produces main.cmo
$ ocamlc -o theprogram Aux.cmo Main.cmo

The program behaves exactly as if the following phrases were entered at top-level:

module Aux: sig (* contents of Aux.mli *) end
          = struct (* contents of Aux.ml *) end;;
module Main: sig (* contents of Main.mli *) end
           = struct (* contents of Main.ml *) end;;

In particular, Main can refer to Aux: the definitions and declarations contained in Main.ml and Main.mli can refer to definition in Aux.ml, using the Aux.ident notation, provided these definitions are exported in Aux.mli.

The order in which the .cmo files are given to ocamlc during the linking phase determines the order in which the module definitions occur. Hence, in the example above, Aux appears first and Main can refer to it, but Aux cannot refer to Main.

Note that only top-level structures can be mapped to separately-compiled files, but neither functors nor module types. However, all module-class objects can appear as components of a structure, so the solution is to put the functor or module type inside a structure, which can then be mapped to a file.

Chapter 3 Objects in OCaml

(Chapter written by JĂ©rĂŽme Vouillon, Didier RĂ©my and Jacques Garrigue)



This chapter gives an overview of the object-oriented features of OCaml.

Note that the relationship between object, class and type in OCaml is different than in mainstream object-oriented languages such as Java and C++, so you shouldn’t assume that similar keywords mean the same thing. Object-oriented features are used much less frequently in OCaml than in those languages. OCaml has alternatives that are often more appropriate, such as modules and functors. Indeed, many OCaml programs do not use objects at all.

3.1 Classes and objects

The class point below defines one instance variable x and two methods get_x and move. The initial value of the instance variable is 0. The variable x is declared mutable, so the method move can change its value.

# class point = object val mutable x = 0 method get_x = x method move d = x <- x + d end;;
class point : object val mutable x : int method get_x : int method move : int -> unit end

We now create a new point p, instance of the point class.

# let p = new point;;
val p : point = <obj>

Note that the type of p is point. This is an abbreviation automatically defined by the class definition above. It stands for the object type <get_x : int; move : int -> unit>, listing the methods of class point along with their types.

We now invoke some methods of p:

# p#get_x;;
- : int = 0
# p#move 3;;
- : unit = ()
# p#get_x;;
- : int = 3

The evaluation of the body of a class only takes place at object creation time. Therefore, in the following example, the instance variable x is initialized to different values for two different objects.

# let x0 = ref 0;;
val x0 : int ref = {contents = 0}
# class point = object val mutable x = incr x0; !x0 method get_x = x method move d = x <- x + d end;;
class point : object val mutable x : int method get_x : int method move : int -> unit end
# new point#get_x;;
- : int = 1
# new point#get_x;;
- : int = 2

The class point can also be abstracted over the initial values of the x coordinate.

# class point = fun x_init -> object val mutable x = x_init method get_x = x method move d = x <- x + d end;;
class point : int -> object val mutable x : int method get_x : int method move : int -> unit end

Like in function definitions, the definition above can be abbreviated as:

# class point x_init = object val mutable x = x_init method get_x = x method move d = x <- x + d end;;
class point : int -> object val mutable x : int method get_x : int method move : int -> unit end

An instance of the class point is now a function that expects an initial parameter to create a point object:

# new point;;
- : int -> point = <fun>
# let p = new point 7;;
val p : point = <obj>

The parameter x_init is, of course, visible in the whole body of the definition, including methods. For instance, the method get_offset in the class below returns the position of the object relative to its initial position.

# class point x_init = object val mutable x = x_init method get_x = x method get_offset = x - x_init method move d = x <- x + d end;;
class point : int -> object val mutable x : int method get_offset : int method get_x : int method move : int -> unit end

Expressions can be evaluated and bound before defining the object body of the class. This is useful to enforce invariants. For instance, points can be automatically adjusted to the nearest point on a grid, as follows:

# class adjusted_point x_init = let origin = (x_init / 10) * 10 in object val mutable x = origin method get_x = x method get_offset = x - origin method move d = x <- x + d end;;
class adjusted_point : int -> object val mutable x : int method get_offset : int method get_x : int method move : int -> unit end

(One could also raise an exception if the x_init coordinate is not on the grid.) In fact, the same effect could be obtained here by calling the definition of class point with the value of the origin.

# class adjusted_point x_init = point ((x_init / 10) * 10);;
class adjusted_point : int -> point

An alternate solution would have been to define the adjustment in a special allocation function:

# let new_adjusted_point x_init = new point ((x_init / 10) * 10);;
val new_adjusted_point : int -> point = <fun>

However, the former pattern is generally more appropriate, since the code for adjustment is part of the definition of the class and will be inherited.

This ability provides class constructors as can be found in other languages. Several constructors can be defined this way to build objects of the same class but with different initialization patterns; an alternative is to use initializers, as described below in section 3.4.

3.2 Immediate objects

There is another, more direct way to create an object: create it without going through a class.

The syntax is exactly the same as for class expressions, but the result is a single object rather than a class. All the constructs described in the rest of this section also apply to immediate objects.

# let p = object val mutable x = 0 method get_x = x method move d = x <- x + d end;;
val p : < get_x : int; move : int -> unit > = <obj>
# p#get_x;;
- : int = 0
# p#move 3;;
- : unit = ()
# p#get_x;;
- : int = 3

Unlike classes, which cannot be defined inside an expression, immediate objects can appear anywhere, using variables from their environment.

# let minmax x y = if x < y then object method min = x method max = y end else object method min = y method max = x end;;
val minmax : 'a -> 'a -> < max : 'a; min : 'a > = <fun>

Immediate objects have two weaknesses compared to classes: their types are not abbreviated, and you cannot inherit from them. But these two weaknesses can be advantages in some situations, as we will see in sections 3.3 and 3.10.

3.3 Reference to self

A method or an initializer can invoke methods on self (that is, the current object). For that, self must be explicitly bound, here to the variable s (s could be any identifier, even though we will often choose the name self.)

# class printable_point x_init = object (s) val mutable x = x_init method get_x = x method move d = x <- x + d method print = print_int s#get_x end;;
class printable_point : int -> object val mutable x : int method get_x : int method move : int -> unit method print : unit end
# let p = new printable_point 7;;
val p : printable_point = <obj>
# p#print;;
7- : unit = ()

Dynamically, the variable s is bound at the invocation of a method. In particular, when the class printable_point is inherited, the variable s will be correctly bound to the object of the subclass.

A common problem with self is that, as its type may be extended in subclasses, you cannot fix it in advance. Here is a simple example.

# let ints = ref [];;
val ints : '_weak1 list ref = {contents = []}
# class my_int = object (self) method n = 1 method register = ints := self :: !ints end ;;
Error: This expression has type < n : int; register : 'a; .. > but an expression was expected of type 'weak1 Self type cannot escape its class

You can ignore the first two lines of the error message. What matters is the last one: putting self into an external reference would make it impossible to extend it through inheritance. We will see in section 3.12 a workaround to this problem. Note however that, since immediate objects are not extensible, the problem does not occur with them.

# let my_int = object (self) method n = 1 method register = ints := self :: !ints end;;
val my_int : < n : int; register : unit > = <obj>

3.4 Initializers

Let-bindings within class definitions are evaluated before the object is constructed. It is also possible to evaluate an expression immediately after the object has been built. Such code is written as an anonymous hidden method called an initializer. Therefore, it can access self and the instance variables.

# class printable_point x_init = let origin = (x_init / 10) * 10 in object (self) val mutable x = origin method get_x = x method move d = x <- x + d method print = print_int self#get_x initializer print_string "new point at "; self#print; print_newline () end;;
class printable_point : int -> object val mutable x : int method get_x : int method move : int -> unit method print : unit end
# let p = new printable_point 17;;
new point at 10 val p : printable_point = <obj>

Initializers cannot be overridden. On the contrary, all initializers are evaluated sequentially. Initializers are particularly useful to enforce invariants. Another example can be seen in section 8.1.

3.5 Virtual methods

It is possible to declare a method without actually defining it, using the keyword virtual. This method will be provided later in subclasses. A class containing virtual methods must be flagged virtual, and cannot be instantiated (that is, no object of this class can be created). It still defines type abbreviations (treating virtual methods as other methods.)

# class virtual abstract_point x_init = object (self) method virtual get_x : int method get_offset = self#get_x - x_init method virtual move : int -> unit end;;
class virtual abstract_point : int -> object method get_offset : int method virtual get_x : int method virtual move : int -> unit end
# class point x_init = object inherit abstract_point x_init val mutable x = x_init method get_x = x method move d = x <- x + d end;;
class point : int -> object val mutable x : int method get_offset : int method get_x : int method move : int -> unit end

Instance variables can also be declared as virtual, with the same effect as with methods.

# class virtual abstract_point2 = object val mutable virtual x : int method move d = x <- x + d end;;
class virtual abstract_point2 : object val mutable virtual x : int method move : int -> unit end
# class point2 x_init = object inherit abstract_point2 val mutable x = x_init method get_offset = x - x_init end;;
class point2 : int -> object val mutable x : int method get_offset : int method move : int -> unit end

3.6 Private methods

Private methods are methods that do not appear in object interfaces. They can only be invoked from other methods of the same object.

# class restricted_point x_init = object (self) val mutable x = x_init method get_x = x method private move d = x <- x + d method bump = self#move 1 end;;
class restricted_point : int -> object val mutable x : int method bump : unit method get_x : int method private move : int -> unit end
# let p = new restricted_point 0;;
val p : restricted_point = <obj>
# p#move 10 ;;
Error: This expression has type restricted_point It has no method move
# p#bump;;
- : unit = ()

Note that this is not the same thing as private and protected methods in Java or C++, which can be called from other objects of the same class. This is a direct consequence of the independence between types and classes in OCaml: two unrelated classes may produce objects of the same type, and there is no way at the type level to ensure that an object comes from a specific class. However a possible encoding of friend methods is given in section 3.17.

Private methods are inherited (they are by default visible in subclasses), unless they are hidden by signature matching, as described below.

Private methods can be made public in a subclass.

# class point_again x = object (self) inherit restricted_point x method virtual move : _ end;;
class point_again : int -> object val mutable x : int method bump : unit method get_x : int method move : int -> unit end

The annotation virtual here is only used to mention a method without providing its definition. Since we didn’t add the private annotation, this makes the method public, keeping the original definition.

An alternative definition is

# class point_again x = object (self : < move : _; ..> ) inherit restricted_point x end;;
class point_again : int -> object val mutable x : int method bump : unit method get_x : int method move : int -> unit end

The constraint on self’s type is requiring a public move method, and this is sufficient to override private.

One could think that a private method should remain private in a subclass. However, since the method is visible in a subclass, it is always possible to pick its code and define a method of the same name that runs that code, so yet another (heavier) solution would be:

# class point_again x = object inherit restricted_point x as super method move = super#move end;;
class point_again : int -> object val mutable x : int method bump : unit method get_x : int method move : int -> unit end

Of course, private methods can also be virtual. Then, the keywords must appear in this order: method private virtual.

3.7 Class interfaces

Class interfaces are inferred from class definitions. They may also be defined directly and used to restrict the type of a class. Like class declarations, they also define a new type abbreviation.

# class type restricted_point_type = object method get_x : int method bump : unit end;;
class type restricted_point_type = object method bump : unit method get_x : int end
# fun (x : restricted_point_type) -> x;;
- : restricted_point_type -> restricted_point_type = <fun>

In addition to program documentation, class interfaces can be used to constrain the type of a class. Both concrete instance variables and concrete private methods can be hidden by a class type constraint. Public methods and virtual members, however, cannot.

# class restricted_point' x = (restricted_point x : restricted_point_type);;
class restricted_point' : int -> restricted_point_type

Or, equivalently:

# class restricted_point' = (restricted_point : int -> restricted_point_type);;
class restricted_point' : int -> restricted_point_type

The interface of a class can also be specified in a module signature, and used to restrict the inferred signature of a module.

# module type POINT = sig class restricted_point' : int -> object method get_x : int method bump : unit end end;;
module type POINT = sig class restricted_point' : int -> object method bump : unit method get_x : int end end
# module Point : POINT = struct class restricted_point' = restricted_point end;;
module Point : POINT

3.8 Inheritance

We illustrate inheritance by defining a class of colored points that inherits from the class of points. This class has all instance variables and all methods of class point, plus a new instance variable c and a new method color.

# class colored_point x (c : string) = object inherit point x val c = c method color = c end;;
class colored_point : int -> string -> object val c : string val mutable x : int method color : string method get_offset : int method get_x : int method move : int -> unit end
# let p' = new colored_point 5 "red";;
val p' : colored_point = <obj>
# p'#get_x, p'#color;;
- : int * string = (5, "red")

A point and a colored point have incompatible types, since a point has no method color. However, the function get_x below is a generic function applying method get_x to any object p that has this method (and possibly some others, which are represented by an ellipsis in the type). Thus, it applies to both points and colored points.

# let get_succ_x p = p#get_x + 1;;
val get_succ_x : < get_x : int; .. > -> int = <fun>
# get_succ_x p + get_succ_x p';;
- : int = 8

Methods need not be declared previously, as shown by the example:

# let set_x p = p#set_x;;
val set_x : < set_x : 'a; .. > -> 'a = <fun>
# let incr p = set_x p (get_succ_x p);;
val incr : < get_x : int; set_x : int -> 'a; .. > -> 'a = <fun>

3.9 Multiple inheritance

Multiple inheritance is allowed. Only the last definition of a method is kept: the redefinition in a subclass of a method that was visible in the parent class overrides the definition in the parent class. Previous definitions of a method can be reused by binding the related ancestor. Below, super is bound to the ancestor printable_point. The name super is a pseudo value identifier that can only be used to invoke a super-class method, as in super#print.

# class printable_colored_point y c = object (self) val c = c method color = c inherit printable_point y as super method! print = print_string "("; super#print; print_string ", "; print_string (self#color); print_string ")" end;;
class printable_colored_point : int -> string -> object val c : string val mutable x : int method color : string method get_x : int method move : int -> unit method print : unit end
# let p' = new printable_colored_point 17 "red";;
new point at (10, red) val p' : printable_colored_point = <obj>
# p'#print;;
(10, red)- : unit = ()

A private method that has been hidden in the parent class is no longer visible, and is thus not overridden. Since initializers are treated as private methods, all initializers along the class hierarchy are evaluated, in the order they are introduced.

Note that for clarity’s sake, the method print is explicitly marked as overriding another definition by annotating the method keyword with an exclamation mark !. If the method print were not overriding the print method of printable_point, the compiler would raise an error:

# object method! m = () end;;
Error: The method `m' has no previous definition

This explicit overriding annotation also works for val and inherit:

# class another_printable_colored_point y c c' = object (self) inherit printable_point y inherit! printable_colored_point y c val! c = c' end;;
class another_printable_colored_point : int -> string -> string -> object val c : string val mutable x : int method color : string method get_x : int method move : int -> unit method print : unit end

3.10 Parameterized classes

Reference cells can be implemented as objects. The naive definition fails to typecheck:

# class oref x_init = object val mutable x = x_init method get = x method set y = x <- y end;;
Error: Some type variables are unbound in this type: class oref : 'a -> object val mutable x : 'a method get : 'a method set : 'a -> unit end The method get has type 'a where 'a is unbound

The reason is that at least one of the methods has a polymorphic type (here, the type of the value stored in the reference cell), thus either the class should be parametric, or the method type should be constrained to a monomorphic type. A monomorphic instance of the class could be defined by:

# class oref (x_init:int) = object val mutable x = x_init method get = x method set y = x <- y end;;
class oref : int -> object val mutable x : int method get : int method set : int -> unit end

Note that since immediate objects do not define a class type, they have no such restriction.

# let new_oref x_init = object val mutable x = x_init method get = x method set y = x <- y end;;
val new_oref : 'a -> < get : 'a; set : 'a -> unit > = <fun>

On the other hand, a class for polymorphic references must explicitly list the type parameters in its declaration. Class type parameters are listed between [ and ]. The type parameters must also be bound somewhere in the class body by a type constraint.

# class ['a] oref x_init = object val mutable x = (x_init : 'a) method get = x method set y = x <- y end;;
class ['a] oref : 'a -> object val mutable x : 'a method get : 'a method set : 'a -> unit end
# let r = new oref 1 in r#set 2; (r#get);;
- : int = 2

The type parameter in the declaration may actually be constrained in the body of the class definition. In the class type, the actual value of the type parameter is displayed in the constraint clause.

# class ['a] oref_succ (x_init:'a) = object val mutable x = x_init + 1 method get = x method set y = x <- y end;;
class ['a] oref_succ : 'a -> object constraint 'a = int val mutable x : int method get : int method set : int -> unit end

Let us consider a more complex example: define a circle, whose center may be any kind of point. We put an additional type constraint in method move, since no free variables must remain unaccounted for by the class type parameters.

# class ['a] circle (c : 'a) = object val mutable center = c method center = center method set_center c = center <- c method move = (center#move : int -> unit) end;;
class ['a] circle : 'a -> object constraint 'a = < move : int -> unit; .. > val mutable center : 'a method center : 'a method move : int -> unit method set_center : 'a -> unit end

An alternate definition of circle, using a constraint clause in the class definition, is shown below. The type #point used below in the constraint clause is an abbreviation produced by the definition of class point. This abbreviation unifies with the type of any object belonging to a subclass of class point. It actually expands to < get_x : int; move : int -> unit; .. >. This leads to the following alternate definition of circle, which has slightly stronger constraints on its argument, as we now expect center to have a method get_x.

# class ['a] circle (c : 'a) = object constraint 'a = #point val mutable center = c method center = center method set_center c = center <- c method move = center#move end;;
class ['a] circle : 'a -> object constraint 'a = #point val mutable center : 'a method center : 'a method move : int -> unit method set_center : 'a -> unit end

The class colored_circle is a specialized version of class circle that requires the type of the center to unify with #colored_point, and adds a method color. Note that when specializing a parameterized class, the instance of type parameter must always be explicitly given. It is again written between [ and ].

# class ['a] colored_circle c = object constraint 'a = #colored_point inherit ['a] circle c method color = center#color end;;
class ['a] colored_circle : 'a -> object constraint 'a = #colored_point val mutable center : 'a method center : 'a method color : string method move : int -> unit method set_center : 'a -> unit end

3.11 Polymorphic methods

While parameterized classes may be polymorphic in their contents, they are not enough to allow polymorphism of method use.

A classical example is defining an iterator.

# List.fold_left;;
- : ('acc -> 'a -> 'acc) -> 'acc -> 'a list -> 'acc = <fun>
# class ['a] intlist (l : int list) = object method empty = (l = []) method fold f (accu : 'a) = List.fold_left f accu l end;;
class ['a] intlist : int list -> object method empty : bool method fold : ('a -> int -> 'a) -> 'a -> 'a end

At first look, we seem to have a polymorphic iterator, however this does not work in practice.

# let l = new intlist [1; 2; 3];;
val l : '_weak2 intlist = <obj>
# l#fold (fun x y -> x+y) 0;;
- : int = 6
# l;;
- : int intlist = <obj>
# l#fold (fun s x -> s ^ Int.to_string x ^ " ") "" ;;
Error: This expression has type int but an expression was expected of type string

Our iterator works, as shows its first use for summation. However, since objects themselves are not polymorphic (only their constructors are), using the fold method fixes its type for this individual object. Our next attempt to use it as a string iterator fails.

The problem here is that quantification was wrongly located: it is not the class we want to be polymorphic, but the fold method. This can be achieved by giving an explicitly polymorphic type in the method definition.

# class intlist (l : int list) = object method empty = (l = []) method fold : 'a. ('a -> int -> 'a) -> 'a -> 'a = fun f accu -> List.fold_left f accu l end;;
class intlist : int list -> object method empty : bool method fold : ('a -> int -> 'a) -> 'a -> 'a end
# let l = new intlist [1; 2; 3];;
val l : intlist = <obj>
# l#fold (fun x y -> x+y) 0;;
- : int = 6
# l#fold (fun s x -> s ^ Int.to_string x ^ " ") "";;
- : string = "1 2 3 "

As you can see in the class type shown by the compiler, while polymorphic method types must be fully explicit in class definitions (appearing immediately after the method name), quantified type variables can be left implicit in class descriptions. Why require types to be explicit? The problem is that (int -> int -> int) -> int -> int would also be a valid type for fold, and it happens to be incompatible with the polymorphic type we gave (automatic instantiation only works for toplevel types variables, not for inner quantifiers, where it becomes an undecidable problem.) So the compiler cannot choose between those two types, and must be helped.

However, the type can be completely omitted in the class definition if it is already known, through inheritance or type constraints on self. Here is an example of method overriding.

# class intlist_rev l = object inherit intlist l method! fold f accu = List.fold_left f accu (List.rev l) end;;

The following idiom separates description and definition.

# class type ['a] iterator = object method fold : ('b -> 'a -> 'b) -> 'b -> 'b end;;
# class intlist' l = object (self : int #iterator) method empty = (l = []) method fold f accu = List.fold_left f accu l end;;

Note here the (self : int #iterator) idiom, which ensures that this object implements the interface iterator.

Polymorphic methods are called in exactly the same way as normal methods, but you should be aware of some limitations of type inference. Namely, a polymorphic method can only be called if its type is known at the call site. Otherwise, the method will be assumed to be monomorphic, and given an incompatible type.

# let sum lst = lst#fold (fun x y -> x+y) 0;;
val sum : < fold : (int -> int -> int) -> int -> 'a; .. > -> 'a = <fun>
# sum l ;;
Error: This expression has type intlist but an expression was expected of type < fold : (int -> int -> int) -> int -> 'a; .. > Types for method fold are incompatible

The workaround is easy: you should put a type constraint on the parameter.

# let sum (lst : _ #iterator) = lst#fold (fun x y -> x+y) 0;;
val sum : int #iterator -> int = <fun>

Of course the constraint may also be an explicit method type. Only occurrences of quantified variables are required.

# let sum lst = (lst : < fold : 'a. ('a -> _ -> 'a) -> 'a -> 'a; .. >)#fold (+) 0;;
val sum : < fold : 'a. ('a -> int -> 'a) -> 'a -> 'a; .. > -> int = <fun>

Another use of polymorphic methods is to allow some form of implicit subtyping in method arguments. We have already seen in section 3.8 how some functions may be polymorphic in the class of their argument. This can be extended to methods.

# class type point0 = object method get_x : int end;;
class type point0 = object method get_x : int end
# class distance_point x = object inherit point x method distance : 'a. (#point0 as 'a) -> int = fun other -> abs (other#get_x - x) end;;
class distance_point : int -> object val mutable x : int method distance : #point0 -> int method get_offset : int method get_x : int method move : int -> unit end
# let p = new distance_point 3 in (p#distance (new point 8), p#distance (new colored_point 1 "blue"));;
- : int * int = (5, 2)

Note here the special syntax (#point0 as 'a) we have to use to quantify the extensible part of #point0. As for the variable binder, it can be omitted in class specifications. If you want polymorphism inside object field it must be quantified independently.

# class multi_poly = object method m1 : 'a. (< n1 : 'b. 'b -> 'b; .. > as 'a) -> _ = fun o -> o#n1 true, o#n1 "hello" method m2 : 'a 'b. (< n2 : 'b -> bool; .. > as 'a) -> 'b -> _ = fun o x -> o#n2 x end;;
class multi_poly : object method m1 : < n1 : 'b. 'b -> 'b; .. > -> bool * string method m2 : < n2 : 'b -> bool; .. > -> 'b -> bool end

In method m1, o must be an object with at least a method n1, itself polymorphic. In method m2, the argument of n2 and x must have the same type, which is quantified at the same level as 'a.

3.12 Using coercions

Subtyping is never implicit. There are, however, two ways to perform subtyping. The most general construction is fully explicit: both the domain and the codomain of the type coercion must be given.

We have seen that points and colored points have incompatible types. For instance, they cannot be mixed in the same list. However, a colored point can be coerced to a point, hiding its color method:

# let colored_point_to_point cp = (cp : colored_point :> point);;
val colored_point_to_point : colored_point -> point = <fun>
# let p = new point 3 and q = new colored_point 4 "blue";;
val p : point = <obj> val q : colored_point = <obj>
# let l = [p; (colored_point_to_point q)];;
val l : point list = [<obj>; <obj>]

An object of type t can be seen as an object of type t' only if t is a subtype of t'. For instance, a point cannot be seen as a colored point.

# (p : point :> colored_point);;
Error: Type point = < get_offset : int; get_x : int; move : int -> unit > is not a subtype of colored_point = < color : string; get_offset : int; get_x : int; move : int -> unit > The first object type has no method color

Indeed, narrowing coercions without runtime checks would be unsafe. Runtime type checks might raise exceptions, and they would require the presence of type information at runtime, which is not the case in the OCaml system. For these reasons, there is no such operation available in the language.

Be aware that subtyping and inheritance are not related. Inheritance is a syntactic relation between classes while subtyping is a semantic relation between types. For instance, the class of colored points could have been defined directly, without inheriting from the class of points; the type of colored points would remain unchanged and thus still be a subtype of points.

The domain of a coercion can often be omitted. For instance, one can define:

# let to_point cp = (cp :> point);;
val to_point : #point -> point = <fun>

In this case, the function colored_point_to_point is an instance of the function to_point. This is not always true, however. The fully explicit coercion is more precise and is sometimes unavoidable. Consider, for example, the following class:

# class c0 = object method m = {< >} method n = 0 end;;
class c0 : object ('a) method m : 'a method n : int end

The object type c0 is an abbreviation for <m : 'a; n : int> as 'a. Consider now the type declaration:

# class type c1 = object method m : c1 end;;
class type c1 = object method m : c1 end

The object type c1 is an abbreviation for the type <m : 'a> as 'a. The coercion from an object of type c0 to an object of type c1 is correct:

# fun (x:c0) -> (x : c0 :> c1);;
- : c0 -> c1 = <fun>

However, the domain of the coercion cannot always be omitted. In that case, the solution is to use the explicit form. Sometimes, a change in the class-type definition can also solve the problem

# class type c2 = object ('a) method m : 'a end;;
class type c2 = object ('a) method m : 'a end
# fun (x:c0) -> (x :> c2);;
- : c0 -> c2 = <fun>

While class types c1 and c2 are different, both object types c1 and c2 expand to the same object type (same method names and types). Yet, when the domain of a coercion is left implicit and its co-domain is an abbreviation of a known class type, then the class type, rather than the object type, is used to derive the coercion function. This allows leaving the domain implicit in most cases when coercing from a subclass to its superclass. The type of a coercion can always be seen as below:

# let to_c1 x = (x :> c1);;
val to_c1 : < m : #c1; .. > -> c1 = <fun>
# let to_c2 x = (x :> c2);;
val to_c2 : #c2 -> c2 = <fun>

Note the difference between these two coercions: in the case of to_c2, the type #c2 = < m : 'a; .. > as 'a is polymorphically recursive (according to the explicit recursion in the class type of c2); hence the success of applying this coercion to an object of class c0. On the other hand, in the first case, c1 was only expanded and unrolled twice to obtain < m : < m : c1; .. >; .. > (remember #c1 = < m : c1; .. >), without introducing recursion. You may also note that the type of to_c2 is #c2 -> c2 while the type of to_c1 is more general than #c1 -> c1. This is not always true, since there are class types for which some instances of #c are not subtypes of c, as explained in section 3.16. Yet, for parameterless classes the coercion (_ :> c) is always more general than (_ : #c :> c).

A common problem may occur when one tries to define a coercion to a class c while defining class c. The problem is due to the type abbreviation not being completely defined yet, and so its subtypes are not clearly known. Then, a coercion (_ :> c) or (_ : #c :> c) is taken to be the identity function, as in

# fun x -> (x :> 'a);;
- : 'a -> 'a = <fun>

As a consequence, if the coercion is applied to self, as in the following example, the type of self is unified with the closed type c (a closed object type is an object type without ellipsis). This would constrain the type of self be closed and is thus rejected. Indeed, the type of self cannot be closed: this would prevent any further extension of the class. Therefore, a type error is generated when the unification of this type with another type would result in a closed object type.

# class c = object method m = 1 end and d = object (self) inherit c method n = 2 method as_c = (self :> c) end;;
Error: This expression cannot be coerced to type c = < m : int >; it has type < as_c : c; m : int; n : int; .. > but is here used with type c Self type cannot escape its class

However, the most common instance of this problem, coercing self to its current class, is detected as a special case by the type checker, and properly typed.

# class c = object (self) method m = (self :> c) end;;
class c : object method m : c end

This allows the following idiom, keeping a list of all objects belonging to a class or its subclasses:

# let all_c = ref [];;
val all_c : '_weak3 list ref = {contents = []}
# class c (m : int) = object (self) method m = m initializer all_c := (self :> c) :: !all_c end;;
class c : int -> object method m : int end

This idiom can in turn be used to retrieve an object whose type has been weakened:

# let rec lookup_obj obj = function [] -> raise Not_found | obj' :: l -> if (obj :> < >) = (obj' :> < >) then obj' else lookup_obj obj l ;;
val lookup_obj : < .. > -> (< .. > as 'a) list -> 'a = <fun>
# let lookup_c obj = lookup_obj obj !all_c;;
val lookup_c : < .. > -> < m : int > = <fun>

The type < m : int > we see here is just the expansion of c, due to the use of a reference; we have succeeded in getting back an object of type c.


The previous coercion problem can often be avoided by first defining the abbreviation, using a class type:

# class type c' = object method m : int end;;
class type c' = object method m : int end
# class c : c' = object method m = 1 end and d = object (self) inherit c method n = 2 method as_c = (self :> c') end;;
class c : c' and d : object method as_c : c' method m : int method n : int end

It is also possible to use a virtual class. Inheriting from this class simultaneously forces all methods of c to have the same type as the methods of c'.

# class virtual c' = object method virtual m : int end;;
class virtual c' : object method virtual m : int end
# class c = object (self) inherit c' method m = 1 end;;
class c : object method m : int end

One could think of defining the type abbreviation directly:

# type c' = <m : int>;;

However, the abbreviation #c' cannot be defined directly in a similar way. It can only be defined by a class or a class-type definition. This is because a #-abbreviation carries an implicit anonymous variable .. that cannot be explicitly named. The closer you get to it is:

# type 'a c'_class = 'a constraint 'a = < m : int; .. >;;

with an extra type variable capturing the open object type.

3.13 Functional objects

It is possible to write a version of class point without assignments on the instance variables. The override construct {< ... >} returns a copy of “self” (that is, the current object), possibly changing the value of some instance variables.

# class functional_point y = object val x = y method get_x = x method move d = {< x = x + d >} method move_to x = {< x >} end;;
class functional_point : int -> object ('a) val x : int method get_x : int method move : int -> 'a method move_to : int -> 'a end
# let p = new functional_point 7;;
val p : functional_point = <obj>
# p#get_x;;
- : int = 7
# (p#move 3)#get_x;;
- : int = 10
# (p#move_to 15)#get_x;;
- : int = 15
# p#get_x;;
- : int = 7

As with records, the form {< x >} is an elided version of {< x = x >} which avoids the repetition of the instance variable name. Note that the type abbreviation functional_point is recursive, which can be seen in the class type of functional_point: the type of self is 'a and 'a appears inside the type of the method move.

The above definition of functional_point is not equivalent to the following:

# class bad_functional_point y = object val x = y method get_x = x method move d = new bad_functional_point (x+d) method move_to x = new bad_functional_point x end;;
class bad_functional_point : int -> object val x : int method get_x : int method move : int -> bad_functional_point method move_to : int -> bad_functional_point end

While objects of either class will behave the same, objects of their subclasses will be different. In a subclass of bad_functional_point, the method move will keep returning an object of the parent class. On the contrary, in a subclass of functional_point, the method move will return an object of the subclass.

Functional update is often used in conjunction with binary methods as illustrated in section 8.2.1.

3.14 Cloning objects

Objects can also be cloned, whether they are functional or imperative. The library function Oo.copy makes a shallow copy of an object. That is, it returns a new object that has the same methods and instance variables as its argument. The instance variables are copied but their contents are shared. Assigning a new value to an instance variable of the copy (using a method call) will not affect instance variables of the original, and conversely. A deeper assignment (for example if the instance variable is a reference cell) will of course affect both the original and the copy.

The type of Oo.copy is the following:

# Oo.copy;;
- : (< .. > as 'a) -> 'a = <fun>

The keyword as in that type binds the type variable 'a to the object type < .. >. Therefore, Oo.copy takes an object with any methods (represented by the ellipsis), and returns an object of the same type. The type of Oo.copy is different from type < .. > -> < .. > as each ellipsis represents a different set of methods. Ellipsis actually behaves as a type variable.

# let p = new point 5;;
val p : point = <obj>
# let q = Oo.copy p;;
val q : point = <obj>
# q#move 7; (p#get_x, q#get_x);;
- : int * int = (5, 12)

In fact, Oo.copy p will behave as p#copy assuming that a public method copy with body {< >} has been defined in the class of p.

Objects can be compared using the generic comparison functions = and <>. Two objects are equal if and only if they are physically equal. In particular, an object and its copy are not equal.

# let q = Oo.copy p;;
val q : point = <obj>
# p = q, p = p;;
- : bool * bool = (false, true)

Other generic comparisons such as (<, <=, ...) can also be used on objects. The relation < defines an unspecified but strict ordering on objects. The ordering relationship between two objects is fixed permanently once the two objects have been created, and it is not affected by mutation of fields.

Cloning and override have a non empty intersection. They are interchangeable when used within an object and without overriding any field:

# class copy = object method copy = {< >} end;;
class copy : object ('a) method copy : 'a end
# class copy = object (self) method copy = Oo.copy self end;;
class copy : object ('a) method copy : 'a end

Only the override can be used to actually override fields, and only the Oo.copy primitive can be used externally.

Cloning can also be used to provide facilities for saving and restoring the state of objects.

# class backup = object (self : 'mytype) val mutable copy = None method save = copy <- Some {< copy = None >} method restore = match copy with Some x -> x | None -> self end;;
class backup : object ('a) val mutable copy : 'a option method restore : 'a method save : unit end

The above definition will only backup one level. The backup facility can be added to any class by using multiple inheritance.

# class ['a] backup_ref x = object inherit ['a] oref x inherit backup end;;
class ['a] backup_ref : 'a -> object ('b) val mutable copy : 'b option val mutable x : 'a method get : 'a method restore : 'b method save : unit method set : 'a -> unit end
# let rec get p n = if n = 0 then p # get else get (p # restore) (n-1);;
val get : (< get : 'b; restore : 'a; .. > as 'a) -> int -> 'b = <fun>
# let p = new backup_ref 0 in p # save; p # set 1; p # save; p # set 2; [get p 0; get p 1; get p 2; get p 3; get p 4];;
- : int list = [2; 1; 1; 1; 1]

We can define a variant of backup that retains all copies. (We also add a method clear to manually erase all copies.)

# class backup = object (self : 'mytype) val mutable copy = None method save = copy <- Some {< >} method restore = match copy with Some x -> x | None -> self method clear = copy <- None end;;
class backup : object ('a) val mutable copy : 'a option method clear : unit method restore : 'a method save : unit end
# class ['a] backup_ref x = object inherit ['a] oref x inherit backup end;;
class ['a] backup_ref : 'a -> object ('b) val mutable copy : 'b option val mutable x : 'a method clear : unit method get : 'a method restore : 'b method save : unit method set : 'a -> unit end
# let p = new backup_ref 0 in p # save; p # set 1; p # save; p # set 2; [get p 0; get p 1; get p 2; get p 3; get p 4];;
- : int list = [2; 1; 0; 0; 0]

3.15 Recursive classes

Recursive classes can be used to define objects whose types are mutually recursive.

# class window = object val mutable top_widget = (None : widget option) method top_widget = top_widget end and widget (w : window) = object val window = w method window = window end;;
class window : object val mutable top_widget : widget option method top_widget : widget option end and widget : window -> object val window : window method window : window end

Although their types are mutually recursive, the classes widget and window are themselves independent.

3.16 Binary methods

A binary method is a method which takes an argument of the same type as self. The class comparable below is a template for classes with a binary method leq of type 'a -> bool where the type variable 'a is bound to the type of self. Therefore, #comparable expands to < leq : 'a -> bool; .. > as 'a. We see here that the binder as also allows writing recursive types.

# class virtual comparable = object (_ : 'a) method virtual leq : 'a -> bool end;;
class virtual comparable : object ('a) method virtual leq : 'a -> bool end

We then define a subclass money of comparable. The class money simply wraps floats as comparable objects.1 We will extend money below with more operations. We have to use a type constraint on the class parameter x because the primitive <= is a polymorphic function in OCaml. The inherit clause ensures that the type of objects of this class is an instance of #comparable.

# class money (x : float) = object inherit comparable val repr = x method value = repr method leq p = repr <= p#value end;;
class money : float -> object ('a) val repr : float method leq : 'a -> bool method value : float end

Note that the type money is not a subtype of type comparable, as the self type appears in contravariant position in the type of method leq. Indeed, an object m of class money has a method leq that expects an argument of type money since it accesses its value method. Considering m of type comparable would allow a call to method leq on m with an argument that does not have a method value, which would be an error.

Similarly, the type money2 below is not a subtype of type money.

# class money2 x = object inherit money x method times k = {< repr = k *. repr >} end;;
class money2 : float -> object ('a) val repr : float method leq : 'a -> bool method times : float -> 'a method value : float end

It is however possible to define functions that manipulate objects of type either money or money2: the function min will return the minimum of any two objects whose type unifies with #comparable. The type of min is not the same as #comparable -> #comparable -> #comparable, as the abbreviation #comparable hides a type variable (an ellipsis). Each occurrence of this abbreviation generates a new variable.

# let min (x : #comparable) y = if x#leq y then x else y;;
val min : (#comparable as 'a) -> 'a -> 'a = <fun>

This function can be applied to objects of type money or money2.

# (min (new money 1.3) (new money 3.1))#value;;
- : float = 1.3
# (min (new money2 5.0) (new money2 3.14))#value;;
- : float = 3.14

More examples of binary methods can be found in sections 8.2.1 and 8.2.3.

Note the use of override for method times. Writing new money2 (k *. repr) instead of {< repr = k *. repr >} would not behave well with inheritance: in a subclass money3 of money2 the times method would return an object of class money2 but not of class money3 as would be expected.

The class money could naturally carry another binary method. Here is a direct definition:

# class money x = object (self : 'a) val repr = x method value = repr method print = print_float repr method times k = {< repr = k *. x >} method leq (p : 'a) = repr <= p#value method plus (p : 'a) = {< repr = x +. p#value >} end;;
class money : float -> object ('a) val repr : float method leq : 'a -> bool method plus : 'a -> 'a method print : unit method times : float -> 'a method value : float end

3.17 Friends

The above class money reveals a problem that often occurs with binary methods. In order to interact with other objects of the same class, the representation of money objects must be revealed, using a method such as value. If we remove all binary methods (here plus and leq), the representation can easily be hidden inside objects by removing the method value as well. However, this is not possible as soon as some binary method requires access to the representation of objects of the same class (other than self).

# class safe_money x = object (self : 'a) val repr = x method print = print_float repr method times k = {< repr = k *. x >} end;;
class safe_money : float -> object ('a) val repr : float method print : unit method times : float -> 'a end

Here, the representation of the object is known only to a particular object. To make it available to other objects of the same class, we are forced to make it available to the whole world. However we can easily restrict the visibility of the representation using the module system.

# module type MONEY = sig type t class c : float -> object ('a) val repr : t method value : t method print : unit method times : float -> 'a method leq : 'a -> bool method plus : 'a -> 'a end end;;
# module Euro : MONEY = struct type t = float class c x = object (self : 'a) val repr = x method value = repr method print = print_float repr method times k = {< repr = k *. x >} method leq (p : 'a) = repr <= p#value method plus (p : 'a) = {< repr = x +. p#value >} end end;;

Another example of friend functions may be found in section 8.2.3. These examples occur when a group of objects (here objects of the same class) and functions should see each others internal representation, while their representation should be hidden from the outside. The solution is always to define all friends in the same module, give access to the representation and use a signature constraint to make the representation abstract outside the module.


1
floats are an approximation of decimal numbers, they are unsuitable for use in most monetary calculations as they may introduce errors.

Chapter 4 Labeled arguments

(Chapter written by Jacques Garrigue)

If you have a look at modules ending in Labels in the standard library, you will see that function types have annotations you did not have in the functions you defined yourself.

# ListLabels.map;;
- : f:('a -> 'b) -> 'a list -> 'b list = <fun>
# StringLabels.sub;;
- : string -> pos:int -> len:int -> string = <fun>

Such annotations of the form name: are called labels. They are meant to document the code, allow more checking, and give more flexibility to function application. You can give such names to arguments in your programs, by prefixing them with a tilde ~.

# let f ~x ~y = x - y;;
val f : x:int -> y:int -> int = <fun>
# let x = 3 and y = 2 in f ~x ~y;;
- : int = 1

When you want to use distinct names for the variable and the label appearing in the type, you can use a naming label of the form ~name:. This also applies when the argument is not a variable.

# let f ~x:x1 ~y:y1 = x1 - y1;;
val f : x:int -> y:int -> int = <fun>
# f ~x:3 ~y:2;;
- : int = 1

Labels obey the same rules as other identifiers in OCaml, that is you cannot use a reserved keyword (like in or to) as a label.

Formal parameters and arguments are matched according to their respective labels, the absence of label being interpreted as the empty label. This allows commuting arguments in applications. One can also partially apply a function on any argument, creating a new function of the remaining parameters.

# let f ~x ~y = x - y;;
val f : x:int -> y:int -> int = <fun>
# f ~y:2 ~x:3;;
- : int = 1
# ListLabels.fold_left;;
- : f:('acc -> 'a -> 'acc) -> init:'acc -> 'a list -> 'acc = <fun>
# ListLabels.fold_left [1;2;3] ~init:0 ~f:( + );;
- : int = 6
# ListLabels.fold_left ~init:0;;
- : f:(int -> 'a -> int) -> 'a list -> int = <fun>

If several arguments of a function bear the same label (or no label), they will not commute among themselves, and order matters. But they can still commute with other arguments.

# let hline ~x:x1 ~x:x2 ~y = (x1, x2, y);;
val hline : x:'a -> x:'b -> y:'c -> 'a * 'b * 'c = <fun>
# hline ~x:3 ~y:2 ~x:5;;
- : int * int * int = (3, 5, 2)

4.1 Optional arguments

An interesting feature of labeled arguments is that they can be made optional. For optional parameters, the question mark ? replaces the tilde ~ of non-optional ones, and the label is also prefixed by ? in the function type. Default values may be given for such optional parameters.

# let bump ?(step = 1) x = x + step;;
val bump : ?step:int -> int -> int = <fun>
# bump 2;;
- : int = 3
# bump ~step:3 2;;
- : int = 5

A function taking some optional arguments must also take at least one non-optional argument. The criterion for deciding whether an optional argument has been omitted is the non-labeled application of an argument appearing after this optional argument in the function type. Note that if that argument is labeled, you will only be able to eliminate optional arguments by totally applying the function, omitting all optional arguments and omitting all labels for all remaining arguments.

# let test ?(x = 0) ?(y = 0) () ?(z = 0) () = (x, y, z);;
val test : ?x:int -> ?y:int -> unit -> ?z:int -> unit -> int * int * int = <fun>
# test ();;
- : ?z:int -> unit -> int * int * int = <fun>
# test ~x:2 () ~z:3 ();;
- : int * int * int = (2, 0, 3)

Optional parameters may also commute with non-optional or unlabeled ones, as long as they are applied simultaneously. By nature, optional arguments do not commute with unlabeled arguments applied independently.

# test ~y:2 ~x:3 () ();;
- : int * int * int = (3, 2, 0)
# test () () ~z:1 ~y:2 ~x:3;;
- : int * int * int = (3, 2, 1)
# (test () ()) ~z:1 ;;
Error: This expression has type int * int * int This is not a function; it cannot be applied.

Here (test () ()) is already (0,0,0) and cannot be further applied.

Optional arguments are actually implemented as option types. If you do not give a default value, you have access to their internal representation, type 'a option = None | Some of 'a. You can then provide different behaviors when an argument is present or not.

# let bump ?step x = match step with | None -> x * 2 | Some y -> x + y ;;
val bump : ?step:int -> int -> int = <fun>

It may also be useful to relay an optional argument from a function call to another. This can be done by prefixing the applied argument with ?. This question mark disables the wrapping of optional argument in an option type.

# let test2 ?x ?y () = test ?x ?y () ();;
val test2 : ?x:int -> ?y:int -> unit -> int * int * int = <fun>
# test2 ?x:None;;
- : ?y:int -> unit -> int * int * int = <fun>

4.2 Labels and type inference

While they provide an increased comfort for writing function applications, labels and optional arguments have the pitfall that they cannot be inferred as completely as the rest of the language.

You can see it in the following two examples.

# let h' g = g ~y:2 ~x:3;;
val h' : (y:int -> x:int -> 'a) -> 'a = <fun>
# h' f ;;
Error: This expression has type x:int -> y:int -> int but an expression was expected of type y:int -> x:int -> 'a
# let bump_it bump x = bump ~step:2 x;;
val bump_it : (step:int -> 'a -> 'b) -> 'a -> 'b = <fun>
# bump_it bump 1 ;;
Error: This expression has type ?step:int -> int -> int but an expression was expected of type step:int -> 'a -> 'b

The first case is simple: g is passed ~y and then ~x, but f expects ~x and then ~y. This is correctly handled if we know the type of g to be x:int -> y:int -> int in advance, but otherwise this causes the above type clash. The simplest workaround is to apply formal parameters in a standard order.

The second example is more subtle: while we intended the argument bump to be of type ?step:int -> int -> int, it is inferred as step:int -> int -> 'a. These two types being incompatible (internally normal and optional arguments are different), a type error occurs when applying bump_it to the real bump.

We will not try here to explain in detail how type inference works. One must just understand that there is not enough information in the above program to deduce the correct type of g or bump. That is, there is no way to know whether an argument is optional or not, or which is the correct order, by looking only at how a function is applied. The strategy used by the compiler is to assume that there are no optional arguments, and that applications are done in the right order.

The right way to solve this problem for optional parameters is to add a type annotation to the argument bump.

# let bump_it (bump : ?step:int -> int -> int) x = bump ~step:2 x;;
val bump_it : (?step:int -> int -> int) -> int -> int = <fun>
# bump_it bump 1;;
- : int = 3

In practice, such problems appear mostly when using objects whose methods have optional arguments, so writing the type of object arguments is often a good idea.

Normally the compiler generates a type error if you attempt to pass to a function a parameter whose type is different from the expected one. However, in the specific case where the expected type is a non-labeled function type, and the argument is a function expecting optional parameters, the compiler will attempt to transform the argument to have it match the expected type, by passing None for all optional parameters.

# let twice f (x : int) = f(f x);;
val twice : (int -> int) -> int -> int = <fun>
# twice bump 2;;
- : int = 8

This transformation is coherent with the intended semantics, including side-effects. That is, if the application of optional parameters shall produce side-effects, these are delayed until the received function is really applied to an argument.

4.3 Suggestions for labeling

Like for names, choosing labels for functions is not an easy task. A good labeling is one which

We explain here the rules we applied when labeling OCaml libraries.

To speak in an “object-oriented” way, one can consider that each function has a main argument, its object, and other arguments related with its action, the parameters. To permit the combination of functions through functionals in commuting label mode, the object will not be labeled. Its role is clear from the function itself. The parameters are labeled with names reminding of their nature or their role. The best labels combine nature and role. When this is not possible the role is to be preferred, since the nature will often be given by the type itself. Obscure abbreviations should be avoided.

ListLabels.map : f:('a -> 'b) -> 'a list -> 'b list
UnixLabels.write : file_descr -> buf:bytes -> pos:int -> len:int -> unit

When there are several objects of same nature and role, they are all left unlabeled.

ListLabels.iter2 : f:('a -> 'b -> unit) -> 'a list -> 'b list -> unit

When there is no preferable object, all arguments are labeled.

BytesLabels.blit :
  src:bytes -> src_pos:int -> dst:bytes -> dst_pos:int -> len:int -> unit

However, when there is only one argument, it is often left unlabeled.

BytesLabels.create : int -> bytes

This principle also applies to functions of several arguments whose return type is a type variable, as long as the role of each argument is not ambiguous. Labeling such functions may lead to awkward error messages when one attempts to omit labels in an application, as we have seen with ListLabels.fold_left.

Here are some of the label names you will find throughout the libraries.

LabelMeaning
f:a function to be applied
pos:a position in a string, array or byte sequence
len:a length
buf:a byte sequence or string used as buffer
src:the source of an operation
dst:the destination of an operation
init:the initial value for an iterator
cmp:a comparison function, e.g. Stdlib.compare
mode:an operation mode or a flag list

All these are only suggestions, but keep in mind that the choice of labels is essential for readability. Bizarre choices will make the program harder to maintain.

In the ideal, the right function name with right labels should be enough to understand the function’s meaning. Since one can get this information with OCamlBrowser or the ocaml toplevel, the documentation is only used when a more detailed specification is needed.

Chapter 5 Polymorphic variants

(Chapter written by Jacques Garrigue)

Variants as presented in section 1.4 are a powerful tool to build data structures and algorithms. However they sometimes lack flexibility when used in modular programming. This is due to the fact that every constructor is assigned to a unique type when defined and used. Even if the same name appears in the definition of multiple types, the constructor itself belongs to only one type. Therefore, one cannot decide that a given constructor belongs to multiple types, or consider a value of some type to belong to some other type with more constructors.

With polymorphic variants, this original assumption is removed. That is, a variant tag does not belong to any type in particular, the type system will just check that it is an admissible value according to its use. You need not define a type before using a variant tag. A variant type will be inferred independently for each of its uses.

5.1 Basic use

In programs, polymorphic variants work like usual ones. You just have to prefix their names with a backquote character `.

# [`On; `Off];;
- : [> `Off | `On ] list = [`On; `Off]
# `Number 1;;
- : [> `Number of int ] = `Number 1
# let f = function `On -> 1 | `Off -> 0 | `Number n -> n;;
val f : [< `Number of int | `Off | `On ] -> int = <fun>
# List.map f [`On; `Off];;
- : int list = [1; 0]

[>`Off|`On] list means that to match this list, you should at least be able to match `Off and `On, without argument. [<`On|`Off|`Number of int] means that f may be applied to `Off, `On (both without argument), or `Number n where n is an integer. The > and < inside the variant types show that they may still be refined, either by defining more tags or by allowing less. As such, they contain an implicit type variable. Because each of the variant types appears only once in the whole type, their implicit type variables are not shown.

The above variant types were polymorphic, allowing further refinement. When writing type annotations, one will most often describe fixed variant types, that is types that cannot be refined. This is also the case for type abbreviations. Such types do not contain < or >, but just an enumeration of the tags and their associated types, just like in a normal datatype definition.

# type 'a vlist = [`Nil | `Cons of 'a * 'a vlist];;
type 'a vlist = [ `Cons of 'a * 'a vlist | `Nil ]
# let rec map f : 'a vlist -> 'b vlist = function | `Nil -> `Nil | `Cons(a, l) -> `Cons(f a, map f l) ;;
val map : ('a -> 'b) -> 'a vlist -> 'b vlist = <fun>

5.2 Advanced use

Type-checking polymorphic variants is a subtle thing, and some expressions may result in more complex type information.

# let f = function `A -> `C | `B -> `D | x -> x;;
val f : ([> `A | `B | `C | `D ] as 'a) -> 'a = <fun>
# f `E;;
- : [> `A | `B | `C | `D | `E ] = `E

Here we are seeing two phenomena. First, since this matching is open (the last case catches any tag), we obtain the type [> `A | `B] rather than [< `A | `B] in a closed matching. Then, since x is returned as is, input and return types are identical. The notation as 'a denotes such type sharing. If we apply f to yet another tag `E, it gets added to the list.

# let f1 = function `A x -> x = 1 | `B -> true | `C -> false let f2 = function `A x -> x = "a" | `B -> true ;;
val f1 : [< `A of int | `B | `C ] -> bool = <fun> val f2 : [< `A of string | `B ] -> bool = <fun>
# let f x = f1 x && f2 x;;
val f : [< `A of string & int | `B ] -> bool = <fun>

Here f1 and f2 both accept the variant tags `A and `B, but the argument of `A is int for f1 and string for f2. In f’s type `C, only accepted by f1, disappears, but both argument types appear for `A as int & string. This means that if we pass the variant tag `A to f, its argument should be both int and string. Since there is no such value, f cannot be applied to `A, and `B is the only accepted input.

Even if a value has a fixed variant type, one can still give it a larger type through coercions. Coercions are normally written with both the source type and the destination type, but in simple cases the source type may be omitted.

# type 'a wlist = [`Nil | `Cons of 'a * 'a wlist | `Snoc of 'a wlist * 'a];;
type 'a wlist = [ `Cons of 'a * 'a wlist | `Nil | `Snoc of 'a wlist * 'a ]
# let wlist_of_vlist l = (l : 'a vlist :> 'a wlist);;
val wlist_of_vlist : 'a vlist -> 'a wlist = <fun>
# let open_vlist l = (l : 'a vlist :> [> 'a vlist]);;
val open_vlist : 'a vlist -> [> 'a vlist ] = <fun>
# fun x -> (x :> [`A|`B|`C]);;
- : [< `A | `B | `C ] -> [ `A | `B | `C ] = <fun>

You may also selectively coerce values through pattern matching.

# let split_cases = function | `Nil | `Cons _ as x -> `A x | `Snoc _ as x -> `B x ;;
val split_cases : [< `Cons of 'a | `Nil | `Snoc of 'b ] -> [> `A of [> `Cons of 'a | `Nil ] | `B of [> `Snoc of 'b ] ] = <fun>

When an or-pattern composed of variant tags is wrapped inside an alias-pattern, the alias is given a type containing only the tags enumerated in the or-pattern. This allows for many useful idioms, like incremental definition of functions.

# let num x = `Num x let eval1 eval (`Num x) = x let rec eval x = eval1 eval x ;;
val num : 'a -> [> `Num of 'a ] = <fun> val eval1 : 'a -> [< `Num of 'b ] -> 'b = <fun> val eval : [< `Num of 'a ] -> 'a = <fun>
# let plus x y = `Plus(x,y) let eval2 eval = function | `Plus(x,y) -> eval x + eval y | `Num _ as x -> eval1 eval x let rec eval x = eval2 eval x ;;
val plus : 'a -> 'b -> [> `Plus of 'a * 'b ] = <fun> val eval2 : ('a -> int) -> [< `Num of int | `Plus of 'a * 'a ] -> int = <fun> val eval : ([< `Num of int | `Plus of 'a * 'a ] as 'a) -> int = <fun>

To make this even more comfortable, you may use type definitions as abbreviations for or-patterns. That is, if you have defined type myvariant = [`Tag1 of int | `Tag2 of bool], then the pattern #myvariant is equivalent to writing (`Tag1(_ : int) | `Tag2(_ : bool)).

Such abbreviations may be used alone,

# let f = function | #myvariant -> "myvariant" | `Tag3 -> "Tag3";;
val f : [< `Tag1 of int | `Tag2 of bool | `Tag3 ] -> string = <fun>

or combined with with aliases.

# let g1 = function `Tag1 _ -> "Tag1" | `Tag2 _ -> "Tag2";;
val g1 : [< `Tag1 of 'a | `Tag2 of 'b ] -> string = <fun>
# let g = function | #myvariant as x -> g1 x | `Tag3 -> "Tag3";;
val g : [< `Tag1 of int | `Tag2 of bool | `Tag3 ] -> string = <fun>

5.3 Weaknesses of polymorphic variants

After seeing the power of polymorphic variants, one may wonder why they were added to core language variants, rather than replacing them.

The answer is twofold. The first aspect is that while being pretty efficient, the lack of static type information allows for less optimizations, and makes polymorphic variants slightly heavier than core language ones. However noticeable differences would only appear on huge data structures.

More important is the fact that polymorphic variants, while being type-safe, result in a weaker type discipline. That is, core language variants do actually much more than ensuring type-safety, they also check that you use only declared constructors, that all constructors present in a data-structure are compatible, and they enforce typing constraints to their parameters.

For this reason, you must be more careful about making types explicit when you use polymorphic variants. When you write a library, this is easy since you can describe exact types in interfaces, but for simple programs you are probably better off with core language variants.

Beware also that some idioms make trivial errors very hard to find. For instance, the following code is probably wrong but the compiler has no way to see it.

# type abc = [`A | `B | `C] ;;
type abc = [ `A | `B | `C ]
# let f = function | `As -> "A" | #abc -> "other" ;;
val f : [< `A | `As | `B | `C ] -> string = <fun>
# let f : abc -> string = f ;;
val f : abc -> string = <fun>

You can avoid such risks by annotating the definition itself.

# let f : abc -> string = function | `As -> "A" | #abc -> "other" ;;
Error: This pattern matches values of type [? `As ] but a pattern was expected which matches values of type abc The second variant type does not allow tag(s) `As

Chapter 6 Polymorphism and its limitations



This chapter covers more advanced questions related to the limitations of polymorphic functions and types. There are some situations in OCaml where the type inferred by the type checker may be less generic than expected. Such non-genericity can stem either from interactions between side-effects and typing or the difficulties of implicit polymorphic recursion and higher-rank polymorphism.

This chapter details each of these situations and, if it is possible, how to recover genericity.

6.1 Weak polymorphism and mutation

6.1.1 Weakly polymorphic types

Maybe the most frequent examples of non-genericity derive from the interactions between polymorphic types and mutation. A simple example appears when typing the following expression

# let store = ref None ;;
val store : '_weak1 option ref = {contents = None}

Since the type of None is 'a option and the function ref has type 'b -> 'b ref, a natural deduction for the type of store would be 'a option ref. However, the inferred type, '_weak1 option ref, is different. Type variables whose names start with a _weak prefix like '_weak1 are weakly polymorphic type variables, sometimes shortened to “weak type variables”. A weak type variable is a placeholder for a single type that is currently unknown. Once the specific type t behind the placeholder type '_weak1 is known, all occurrences of '_weak1 will be replaced by t. For instance, we can define another option reference and store an int inside:

# let another_store = ref None ;;
val another_store : '_weak2 option ref = {contents = None}
# another_store := Some 0; another_store ;;
- : int option ref = {contents = Some 0}

After storing an int inside another_store, the type of another_store has been updated from '_weak2 option ref to int option ref. This distinction between weakly and generic polymorphic type variable protects OCaml programs from unsoundness and runtime errors. To understand from where unsoundness might come, consider this simple function which swaps a value x with the value stored inside a store reference, if there is such value:

# let swap store x = match !store with | None -> store := Some x; x | Some y -> store := Some x; y;;
val swap : 'a option ref -> 'a -> 'a = <fun>

We can apply this function to our store

# let one = swap store 1 let one_again = swap store 2 let two = swap store 3;;
val one : int = 1 val one_again : int = 1 val two : int = 2

After these three swaps the stored value is 3. Everything is fine up to now. We can then try to swap 3 with a more interesting value, for instance a function:

# let error = swap store (fun x -> x);;
Error: This expression should not be a function, the expected type is int

At this point, the type checker rightfully complains that it is not possible to swap an integer and a function, and that an int should always be traded for another int. Furthermore, the type checker prevents us from manually changing the type of the value stored by store:

# store := Some (fun x -> x);;
Error: This expression should not be a function, the expected type is int

Indeed, looking at the type of store, we see that the weak type '_weak1 has been replaced by the type int

# store;;
- : int option ref = {contents = Some 3}

Therefore, after placing an int in store, we cannot use it to store any value other than an int. More generally, weak types protect the program from undue mutation of values with a polymorphic type.

Moreover, weak types cannot appear in the signature of toplevel modules: types must be known at compilation time. Otherwise, different compilation units could replace the weak type with different and incompatible types. For this reason, compiling the following small piece of code

let option_ref = ref None

yields a compilation error

Error: The type of this expression, '_weak1 option ref,
       contains type variables that cannot be generalized

To solve this error, it is enough to add an explicit type annotation to specify the type at declaration time:

let option_ref: int option ref = ref None

This is in any case a good practice for such global mutable variables. Otherwise, they will pick out the type of first use. If there is a mistake at this point, it can result in confusing type errors when later, correct uses are flagged as errors.

6.1.2 The value restriction

Identifying the exact context in which polymorphic types should be replaced by weak types in a modular way is a difficult question. Indeed the type system must handle the possibility that functions may hide persistent mutable states. For instance, the following function uses an internal reference to implement a delayed identity function

# let make_fake_id () = let store = ref None in fun x -> swap store x ;;
val make_fake_id : unit -> 'a -> 'a = <fun>
# let fake_id = make_fake_id();;
val fake_id : '_weak3 -> '_weak3 = <fun>

It would be unsound to apply this fake_id function to values with different types. The function fake_id is therefore rightfully assigned the type '_weak3 -> '_weak3 rather than 'a -> 'a. At the same time, it ought to be possible to use a local mutable state without impacting the type of a function.

To circumvent these dual difficulties, the type checker considers that any value returned by a function might rely on persistent mutable states behind the scene and should be given a weak type. This restriction on the type of mutable values and the results of function application is called the value restriction. Note that this value restriction is conservative: there are situations where the value restriction is too cautious and gives a weak type to a value that could be safely generalized to a polymorphic type:

# let not_id = (fun x -> x) (fun x -> x);;
val not_id : '_weak4 -> '_weak4 = <fun>

Quite often, this happens when defining functions using higher order functions. To avoid this problem, a solution is to add an explicit argument to the function:

# let id_again = fun x -> (fun x -> x) (fun x -> x) x;;
val id_again : 'a -> 'a = <fun>

With this argument, id_again is seen as a function definition by the type checker and can therefore be generalized. This kind of manipulation is called eta-expansion in lambda calculus and is sometimes referred under this name.

6.1.3 The relaxed value restriction

There is another partial solution to the problem of unnecessary weak types, which is implemented directly within the type checker. Briefly, it is possible to prove that weak types that only appear as type parameters in covariant positions –also called positive positions– can be safely generalized to polymorphic types. For instance, the type 'a list is covariant in 'a:

# let f () = [];;
val f : unit -> 'a list = <fun>
# let empty = f ();;
val empty : 'a list = []

Note that the type inferred for empty is 'a list and not the '_weak5 list that should have occurred with the value restriction.

The value restriction combined with this generalization for covariant type parameters is called the relaxed value restriction.

6.1.4 Variance and value restriction

Variance describes how type constructors behave with respect to subtyping. Consider for instance a pair of type x and xy with x a subtype of xy, denoted x :> xy:

# type x = [ `X ];;
type x = [ `X ]
# type xy = [ `X | `Y ];;
type xy = [ `X | `Y ]

As x is a subtype of xy, we can convert a value of type x to a value of type xy:

# let x:x = `X;;
val x : x = `X
# let x' = ( x :> xy);;
val x' : xy = `X

Similarly, if we have a value of type x list, we can convert it to a value of type xy list, since we could convert each element one by one:

# let l:x list = [`X; `X];;
val l : x list = [`X; `X]
# let l' = ( l :> xy list);;
val l' : xy list = [`X; `X]

In other words, x :> xy implies that x list :> xy list, therefore the type constructor 'a list is covariant (it preserves subtyping) in its parameter 'a.

Contrarily, if we have a function that can handle values of type xy

# let f: xy -> unit = function | `X -> () | `Y -> ();;
val f : xy -> unit = <fun>

it can also handle values of type x:

# let f' = (f :> x -> unit);;
val f' : x -> unit = <fun>

Note that we can rewrite the type of f and f' as

# type 'a proc = 'a -> unit let f' = (f: xy proc :> x proc);;
type 'a proc = 'a -> unit val f' : x proc = <fun>

In this case, we have x :> xy implies xy proc :> x proc. Notice that the second subtyping relation reverse the order of x and xy: the type constructor 'a proc is contravariant in its parameter 'a. More generally, the function type constructor 'a -> 'b is covariant in its return type 'b and contravariant in its argument type 'a.

A type constructor can also be invariant in some of its type parameters, neither covariant nor contravariant. A typical example is a reference:

# let x: x ref = ref `X;;
val x : x ref = {contents = `X}

If we were able to coerce x to the type xy ref as a variable xy, we could use xy to store the value `Y inside the reference and then use the x value to read this content as a value of type x, which would break the type system.

More generally, as soon as a type variable appears in a position describing mutable state it becomes invariant. As a corollary, covariant variables will never denote mutable locations and can be safely generalized. For a better description, interested readers can consult the original article by Jacques Garrigue on http://www.math.nagoya-u.ac.jp/~garrigue/papers/morepoly-long.pdf

Together, the relaxed value restriction and type parameter covariance help to avoid eta-expansion in many situations.

6.1.5 Abstract data types

Moreover, when the type definitions are exposed, the type checker is able to infer variance information on its own and one can benefit from the relaxed value restriction even unknowingly. However, this is not the case anymore when defining new abstract types. As an illustration, we can define a module type collection as:

# module type COLLECTION = sig type 'a t val empty: unit -> 'a t end module Implementation = struct type 'a t = 'a list let empty ()= [] end;;
module type COLLECTION = sig type 'a t val empty : unit -> 'a t end module Implementation : sig type 'a t = 'a list val empty : unit -> 'a list end
# module List2: COLLECTION = Implementation;;
module List2 : COLLECTION

In this situation, when coercing the module List2 to the module type COLLECTION, the type checker forgets that 'a List2.t was covariant in 'a. Consequently, the relaxed value restriction does not apply anymore:

# List2.empty ();;
- : '_weak5 List2.t = <abstr>

To keep the relaxed value restriction, we need to declare the abstract type 'a COLLECTION.t as covariant in 'a:

# module type COLLECTION = sig type +'a t val empty: unit -> 'a t end module List2: COLLECTION = Implementation;;
module type COLLECTION = sig type +'a t val empty : unit -> 'a t end module List2 : COLLECTION

We then recover polymorphism:

# List2.empty ();;
- : 'a List2.t = <abstr>

6.2 Polymorphic recursion

The second major class of non-genericity is directly related to the problem of type inference for polymorphic functions. In some circumstances, the type inferred by OCaml might be not general enough to allow the definition of some recursive functions, in particular for recursive functions acting on non-regular algebraic data types.

With a regular polymorphic algebraic data type, the type parameters of the type constructor are constant within the definition of the type. For instance, we can look at arbitrarily nested list defined as:

# type 'a regular_nested = List of 'a list | Nested of 'a regular_nested list let l = Nested[ List [1]; Nested [List[2;3]]; Nested[Nested[]] ];;
type 'a regular_nested = List of 'a list | Nested of 'a regular_nested list val l : int regular_nested = Nested [List [1]; Nested [List [2; 3]]; Nested [Nested []]]

Note that the type constructor regular_nested always appears as 'a regular_nested in the definition above, with the same parameter 'a. Equipped with this type, one can compute a maximal depth with a classic recursive function

# let rec maximal_depth = function | List _ -> 1 | Nested [] -> 0 | Nested (a::q) -> 1 + max (maximal_depth a) (maximal_depth (Nested q));;
val maximal_depth : 'a regular_nested -> int = <fun>

Non-regular recursive algebraic data types correspond to polymorphic algebraic data types whose parameter types vary between the left and right side of the type definition. For instance, it might be interesting to define a datatype that ensures that all lists are nested at the same depth:

# type 'a nested = List of 'a list | Nested of 'a list nested;;
type 'a nested = List of 'a list | Nested of 'a list nested

Intuitively, a value of type 'a nested is a list of list 
of list of elements a with k nested list. We can then adapt the maximal_depth function defined on regular_depth into a depth function that computes this k. As a first try, we may define

# let rec depth = function | List _ -> 1 | Nested n -> 1 + depth n;;
Error: This expression has type 'a list nested but an expression was expected of type 'a nested The type variable 'a occurs inside 'a list

The type error here comes from the fact that during the definition of depth, the type checker first assigns to depth the type 'a -> 'b . When typing the pattern matching, 'a -> 'b becomes 'a nested -> 'b, then 'a nested -> int once the List branch is typed. However, when typing the application depth n in the Nested branch, the type checker encounters a problem: depth n is applied to 'a list nested, it must therefore have the type 'a list nested -> 'b. Unifying this constraint with the previous one leads to the impossible constraint 'a list nested = 'a nested. In other words, within its definition, the recursive function depth is applied to values of type 'a t with different types 'a due to the non-regularity of the type constructor nested. This creates a problem because the type checker had introduced a new type variable 'a only at the definition of the function depth whereas, here, we need a different type variable for every application of the function depth.

6.2.1 Explicitly polymorphic annotations

The solution of this conundrum is to use an explicitly polymorphic type annotation for the type 'a:

# let rec depth: 'a. 'a nested -> int = function | List _ -> 1 | Nested n -> 1 + depth n;;
val depth : 'a nested -> int = <fun>
# depth ( Nested(List [ [7]; [8] ]) );;
- : int = 2

In the type of depth, 'a.'a nested -> int, the type variable 'a is universally quantified. In other words, 'a.'a nested -> int reads as “for all type 'a, depth maps 'a nested values to integers”. Whereas the standard type 'a nested -> int can be interpreted as “let be a type variable 'a, then depth maps 'a nested values to integers”. There are two major differences with these two type expressions. First, the explicit polymorphic annotation indicates to the type checker that it needs to introduce a new type variable every time the function depth is applied. This solves our problem with the definition of the function depth.

Second, it also notifies the type checker that the type of the function should be polymorphic. Indeed, without explicit polymorphic type annotation, the following type annotation is perfectly valid

# let sum: 'a -> 'b -> 'c = fun x y -> x + y;;
val sum : int -> int -> int = <fun>

since 'a,'b and 'c denote type variables that may or may not be polymorphic. Whereas, it is an error to unify an explicitly polymorphic type with a non-polymorphic type:

# let sum: 'a 'b 'c. 'a -> 'b -> 'c = fun x y -> x + y;;
Error: This definition has type int -> int -> int which is less general than 'a 'b 'c. 'a -> 'b -> 'c

An important remark here is that it is not needed to explicit fully the type of depth: it is sufficient to add annotations only for the universally quantified type variables:

# let rec depth: 'a. 'a nested -> _ = function | List _ -> 1 | Nested n -> 1 + depth n;;
val depth : 'a nested -> int = <fun>
# depth ( Nested(List [ [7]; [8] ]) );;
- : int = 2

6.2.2 More examples

With explicit polymorphic annotations, it becomes possible to implement any recursive function that depends only on the structure of the nested lists and not on the type of the elements. For instance, a more complex example would be to compute the total number of elements of the nested lists:

# let len nested = let map_and_sum f = List.fold_left (fun acc x -> acc + f x) 0 in let rec len: 'a. ('a list -> int ) -> 'a nested -> int = fun nested_len n -> match n with | List l -> nested_len l | Nested n -> len (map_and_sum nested_len) n in len List.length nested;;
val len : 'a nested -> int = <fun>
# len (Nested(Nested(List [ [ [1;2]; [3] ]; [ []; [4]; [5;6;7]]; [[]] ])));;
- : int = 7

Similarly, it may be necessary to use more than one explicitly polymorphic type variables, like for computing the nested list of list lengths of the nested list:

# let shape n = let rec shape: 'a 'b. ('a nested -> int nested) -> ('b list list -> 'a list) -> 'b nested -> int nested = fun nest nested_shape -> function | List l -> raise (Invalid_argument "shape requires nested_list of depth greater than 1") | Nested (List l) -> nest @@ List (nested_shape l) | Nested n -> let nested_shape = List.map nested_shape in let nest x = nest (Nested x) in shape nest nested_shape n in shape (fun n -> n ) (fun l -> List.map List.length l ) n;;
val shape : 'a nested -> int nested = <fun>
# shape (Nested(Nested(List [ [ [1;2]; [3] ]; [ []; [4]; [5;6;7]]; [[]] ])));;
- : int nested = Nested (List [[2; 1]; [0; 1; 3]; [0]])

6.3 Higher-rank polymorphic functions

Explicit polymorphic annotations are however not sufficient to cover all the cases where the inferred type of a function is less general than expected. A similar problem arises when using polymorphic functions as arguments of higher-order functions. For instance, we may want to compute the average depth or length of two nested lists:

# let average_depth x y = (depth x + depth y) / 2;;
val average_depth : 'a nested -> 'b nested -> int = <fun>
# let average_len x y = (len x + len y) / 2;;
val average_len : 'a nested -> 'b nested -> int = <fun>
# let one = average_len (List [2]) (List [[]]);;
val one : int = 1

It would be natural to factorize these two definitions as:

# let average f x y = (f x + f y) / 2;;
val average : ('a -> int) -> 'a -> 'a -> int = <fun>

However, the type of average len is less generic than the type of average_len, since it requires the type of the first and second argument to be the same:

# average_len (List [2]) (List [[]]);;
- : int = 1
# average len (List [2]) (List [[]]);;
Error: This expression has type 'a list but an expression was expected of type int

As previously with polymorphic recursion, the problem stems from the fact that type variables are introduced only at the start of the let definitions. When we compute both f x and f y, the type of x and y are unified together. To avoid this unification, we need to indicate to the type checker that f is polymorphic in its first argument. In some sense, we would want average to have type

val average: ('a. 'a nested -> int) -> 'a nested -> 'b nested -> int

Note that this syntax is not valid within OCaml: average has an universally quantified type 'a inside the type of one of its argument whereas for polymorphic recursion the universally quantified type was introduced before the rest of the type. This position of the universally quantified type means that average is a second-rank polymorphic function. This kind of higher-rank functions is not directly supported by OCaml: type inference for second-rank polymorphic function and beyond is undecidable; therefore using this kind of higher-rank functions requires to handle manually these universally quantified types.

In OCaml, there are two ways to introduce this kind of explicit universally quantified types: universally quantified record fields,

# type 'a nested_reduction = { f:'elt. 'elt nested -> 'a };;
type 'a nested_reduction = { f : 'elt. 'elt nested -> 'a; }
# let boxed_len = { f = len };;
val boxed_len : int nested_reduction = {f = <fun>}

and universally quantified object methods:

# let obj_len = object method f:'a. 'a nested -> 'b = len end;;
val obj_len : < f : 'a. 'a nested -> int > = <obj>

To solve our problem, we can therefore use either the record solution:

# let average nsm x y = (nsm.f x + nsm.f y) / 2 ;;
val average : int nested_reduction -> 'a nested -> 'b nested -> int = <fun>

or the object one:

# let average (obj:<f:'a. 'a nested -> _ > ) x y = (obj#f x + obj#f y) / 2 ;;
val average : < f : 'a. 'a nested -> int > -> 'b nested -> 'c nested -> int = <fun>

Chapter 7 Generalized algebraic datatypes

Generalized algebraic datatypes, or GADTs, extend usual sum types in two ways: constraints on type parameters may change depending on the value constructor, and some type variables may be existentially quantified. Adding constraints is done by giving an explicit return type, where type parameters are instantiated:

type _ term = | Int : int -> int term | Add : (int -> int -> int) term | App : ('b -> 'a) term * 'b term -> 'a term

This return type must use the same type constructor as the type being defined, and have the same number of parameters. Variables are made existential when they appear inside a constructor’s argument, but not in its return type. Since the use of a return type often eliminates the need to name type parameters in the left-hand side of a type definition, one can replace them with anonymous types _ in that case.

The constraints associated to each constructor can be recovered through pattern-matching. Namely, if the type of the scrutinee of a pattern-matching contains a locally abstract type, this type can be refined according to the constructor used. These extra constraints are only valid inside the corresponding branch of the pattern-matching. If a constructor has some existential variables, fresh locally abstract types are generated, and they must not escape the scope of this branch.

7.1 Recursive functions

We write an eval function:

let rec eval : type a. a term -> a = function | Int n -> n (* a = int *) | Add -> (fun x y -> x+y) (* a = int -> int -> int *) | App(f,x) -> (eval f) (eval x) (* eval called at types (b->a) and b for fresh b *)

And use it:

let two = eval (App (App (Add, Int 1), Int 1))
val two : int = 2

It is important to remark that the function eval is using the polymorphic syntax for locally abstract types. When defining a recursive function that manipulates a GADT, explicit polymorphic recursion should generally be used. For instance, the following definition fails with a type error:

let rec eval (type a) : a term -> a = function | Int n -> n | Add -> (fun x y -> x+y) | App(f,x) -> (eval f) (eval x)
Error: This expression has type ($App_'b -> a) term but an expression was expected of type 'a The type constructor $App_'b would escape its scope

In absence of an explicit polymorphic annotation, a monomorphic type is inferred for the recursive function. If a recursive call occurs inside the function definition at a type that involves an existential GADT type variable, this variable flows to the type of the recursive function, and thus escapes its scope. In the above example, this happens in the branch App(f,x) when eval is called with f as an argument. In this branch, the type of f is ($App_'b -> a) term. The prefix $ in $App_'b denotes an existential type named by the compiler (see 7.5). Since the type of eval is 'a term -> 'a, the call eval f makes the existential type $App_'b flow to the type variable 'a and escape its scope. This triggers the above error.

7.2 Type inference

Type inference for GADTs is notoriously hard. This is due to the fact some types may become ambiguous when escaping from a branch. For instance, in the Int case above, n could have either type int or a, and they are not equivalent outside of that branch. As a first approximation, type inference will always work if a pattern-matching is annotated with types containing no free type variables (both on the scrutinee and the return type). This is the case in the above example, thanks to the type annotation containing only locally abstract types.

In practice, type inference is a bit more clever than that: type annotations do not need to be immediately on the pattern-matching, and the types do not have to be always closed. As a result, it is usually enough to only annotate functions, as in the example above. Type annotations are propagated in two ways: for the scrutinee, they follow the flow of type inference, in a way similar to polymorphic methods; for the return type, they follow the structure of the program, they are split on functions, propagated to all branches of a pattern matching, and go through tuples, records, and sum types. Moreover, the notion of ambiguity used is stronger: a type is only seen as ambiguous if it was mixed with incompatible types (equated by constraints), without type annotations between them. For instance, the following program types correctly.

let rec sum : type a. a term -> _ = fun x -> let y = match x with | Int n -> n | Add -> 0 | App(f,x) -> sum f + sum x in y + 1
val sum : 'a term -> int = <fun>

Here the return type int is never mixed with a, so it is seen as non-ambiguous, and can be inferred. When using such partial type annotations we strongly suggest specifying the -principal mode, to check that inference is principal.

The exhaustiveness check is aware of GADT constraints, and can automatically infer that some cases cannot happen. For instance, the following pattern matching is correctly seen as exhaustive (the Add case cannot happen).

let get_int : int term -> int = function | Int n -> n | App(_,_) -> 0

7.3 Refutation cases

Usually, the exhaustiveness check only tries to check whether the cases omitted from the pattern matching are typable or not. However, you can force it to try harder by adding refutation cases, written as a full stop. In the presence of a refutation case, the exhaustiveness check will first compute the intersection of the pattern with the complement of the cases preceding it. It then checks whether the resulting patterns can really match any concrete values by trying to type-check them. Wild cards in the generated patterns are handled in a special way: if their type is a variant type with only GADT constructors, then the pattern is split into the different constructors, in order to check whether any of them is possible (this splitting is not done for arguments of these constructors, to avoid non-termination). We also split tuples and variant types with only one case, since they may contain GADTs inside. For instance, the following code is deemed exhaustive:

type _ t = | Int : int t | Bool : bool t let deep : (char t * int) option -> char = function | None -> 'c' | _ -> .

Namely, the inferred remaining case is Some _, which is split into Some (Int, _) and Some (Bool, _), which are both untypable because deep expects a non-existing char t as the first element of the tuple. Note that the refutation case could be omitted here, because it is automatically added when there is only one case in the pattern matching.

Another addition is that the redundancy check is now aware of GADTs: a case will be detected as redundant if it could be replaced by a refutation case using the same pattern.

7.4 Advanced examples

The term type we have defined above is an indexed type, where a type parameter reflects a property of the value contents. Another use of GADTs is singleton types, where a GADT value represents exactly one type. This value can be used as runtime representation for this type, and a function receiving it can have a polytypic behavior.

Here is an example of a polymorphic function that takes the runtime representation of some type t and a value of the same type, then pretty-prints the value as a string:

type _ typ = | Int : int typ | String : string typ | Pair : 'a typ * 'b typ -> ('a * 'b) typ let rec to_string: type t. t typ -> t -> string = fun t x -> match t with | Int -> Int.to_string x | String -> Printf.sprintf "%S" x | Pair(t1,t2) -> let (x1, x2) = x in Printf.sprintf "(%s,%s)" (to_string t1 x1) (to_string t2 x2)

Another frequent application of GADTs is equality witnesses.

type (_,_) eq = Eq : ('a,'a) eq let cast : type a b. (a,b) eq -> a -> b = fun Eq x -> x

Here type eq has only one constructor, and by matching on it one adds a local constraint allowing the conversion between a and b. By building such equality witnesses, one can make equal types which are syntactically different.

Here is an example using both singleton types and equality witnesses to implement dynamic types.

let rec eq_type : type a b. a typ -> b typ -> (a,b) eq option = fun a b -> match a, b with | Int, Int -> Some Eq | String, String -> Some Eq | Pair(a1,a2), Pair(b1,b2) -> begin match eq_type a1 b1, eq_type a2 b2 with | Some Eq, Some Eq -> Some Eq | _ -> None end | _ -> None type dyn = Dyn : 'a typ * 'a -> dyn let get_dyn : type a. a typ -> dyn -> a option = fun a (Dyn(b,x)) -> match eq_type a b with | None -> None | Some Eq -> Some x

7.5 Existential type names in error messages

The typing of pattern matching in the presence of GADTs can generate many existential types. When necessary, error messages refer to these existential types using compiler-generated names. Currently, the compiler generates these names according to the following nomenclature:

As shown by the last item, the current behavior is imperfect and may be improved in future versions.

7.6 Explicit naming of existentials

As explained above, pattern-matching on a GADT constructor may introduce existential types. Syntax has been introduced which allows them to be named explicitly. For instance, the following code names the type of the argument of f and uses this name.

type _ closure = Closure : ('a -> 'b) * 'a -> 'b closure let eval = fun (Closure (type a) (f, x : (a -> _) * _)) -> f (x : a)

All existential type variables of the constructor must by introduced by the (type ...) construct and bound by a type annotation on the outside of the constructor argument.

7.7 Equations on non-local abstract types

GADT pattern-matching may also add type equations to non-local abstract types. The behaviour is the same as with local abstract types. Reusing the above eq type, one can write:

module M : sig type t val x : t val e : (t,int) eq end = struct type t = int let x = 33 let e = Eq end let x : int = let Eq = M.e in M.x

Of course, not all abstract types can be refined, as this would contradict the exhaustiveness check. Namely, builtin types (those defined by the compiler itself, such as int or array), and abstract types defined by the local module, are non-instantiable, and as such cause a type error rather than introduce an equation.

Chapter 8 Advanced examples with classes and modules

(Chapter written by Didier RĂ©my)



In this chapter, we show some larger examples using objects, classes and modules. We review many of the object features simultaneously on the example of a bank account. We show how modules taken from the standard library can be expressed as classes. Lastly, we describe a programming pattern known as virtual types through the example of window managers.

8.1 Extended example: bank accounts

In this section, we illustrate most aspects of Object and inheritance by refining, debugging, and specializing the following initial naive definition of a simple bank account. (We reuse the module Euro defined at the end of chapter 3.)

# let euro = new Euro.c;;
val euro : float -> Euro.c = <fun>
# let zero = euro 0.;;
val zero : Euro.c = <obj>
# let neg x = x#times (-1.);;
val neg : < times : float -> 'a; .. > -> 'a = <fun>
# class account = object val mutable balance = zero method balance = balance method deposit x = balance <- balance # plus x method withdraw x = if x#leq balance then (balance <- balance # plus (neg x); x) else zero end;;
class account : object val mutable balance : Euro.c method balance : Euro.c method deposit : Euro.c -> unit method withdraw : Euro.c -> Euro.c end
# let c = new account in c # deposit (euro 100.); c # withdraw (euro 50.);;
- : Euro.c = <obj>

We now refine this definition with a method to compute interest.

# class account_with_interests = object (self) inherit account method private interest = self # deposit (self # balance # times 0.03) end;;
class account_with_interests : object val mutable balance : Euro.c method balance : Euro.c method deposit : Euro.c -> unit method private interest : unit method withdraw : Euro.c -> Euro.c end

We make the method interest private, since clearly it should not be called freely from the outside. Here, it is only made accessible to subclasses that will manage monthly or yearly updates of the account.

We should soon fix a bug in the current definition: the deposit method can be used for withdrawing money by depositing negative amounts. We can fix this directly:

# class safe_account = object inherit account method deposit x = if zero#leq x then balance <- balance#plus x end;;
class safe_account : object val mutable balance : Euro.c method balance : Euro.c method deposit : Euro.c -> unit method withdraw : Euro.c -> Euro.c end

However, the bug might be fixed more safely by the following definition:

# class safe_account = object inherit account as unsafe method deposit x = if zero#leq x then unsafe # deposit x else raise (Invalid_argument "deposit") end;;
class safe_account : object val mutable balance : Euro.c method balance : Euro.c method deposit : Euro.c -> unit method withdraw : Euro.c -> Euro.c end

In particular, this does not require the knowledge of the implementation of the method deposit.

To keep track of operations, we extend the class with a mutable field history and a private method trace to add an operation in the log. Then each method to be traced is redefined.

# type 'a operation = Deposit of 'a | Retrieval of 'a;;
type 'a operation = Deposit of 'a | Retrieval of 'a
# class account_with_history = object (self) inherit safe_account as super val mutable history = [] method private trace x = history <- x :: history method deposit x = self#trace (Deposit x); super#deposit x method withdraw x = self#trace (Retrieval x); super#withdraw x method history = List.rev history end;;
class account_with_history : object val mutable balance : Euro.c val mutable history : Euro.c operation list method balance : Euro.c method deposit : Euro.c -> unit method history : Euro.c operation list method private trace : Euro.c operation -> unit method withdraw : Euro.c -> Euro.c end

One may wish to open an account and simultaneously deposit some initial amount. Although the initial implementation did not address this requirement, it can be achieved by using an initializer.

# class account_with_deposit x = object inherit account_with_history initializer balance <- x end;;
class account_with_deposit : Euro.c -> object val mutable balance : Euro.c val mutable history : Euro.c operation list method balance : Euro.c method deposit : Euro.c -> unit method history : Euro.c operation list method private trace : Euro.c operation -> unit method withdraw : Euro.c -> Euro.c end

A better alternative is:

# class account_with_deposit x = object (self) inherit account_with_history initializer self#deposit x end;;
class account_with_deposit : Euro.c -> object val mutable balance : Euro.c val mutable history : Euro.c operation list method balance : Euro.c method deposit : Euro.c -> unit method history : Euro.c operation list method private trace : Euro.c operation -> unit method withdraw : Euro.c -> Euro.c end

Indeed, the latter is safer since the call to deposit will automatically benefit from safety checks and from the trace. Let’s test it:

# let ccp = new account_with_deposit (euro 100.) in let _balance = ccp#withdraw (euro 50.) in ccp#history;;
- : Euro.c operation list = [Deposit <obj>; Retrieval <obj>]

Closing an account can be done with the following polymorphic function:

# let close c = c#withdraw c#balance;;
val close : < balance : 'a; withdraw : 'a -> 'b; .. > -> 'b = <fun>

Of course, this applies to all sorts of accounts.

Finally, we gather several versions of the account into a module Account abstracted over some currency.

# let today () = (01,01,2000) (* an approximation *) module Account (M:MONEY) = struct type m = M.c let m = new M.c let zero = m 0. class bank = object (self) val mutable balance = zero method balance = balance val mutable history = [] method private trace x = history <- x::history method deposit x = self#trace (Deposit x); if zero#leq x then balance <- balance # plus x else raise (Invalid_argument "deposit") method withdraw x = if x#leq balance then (balance <- balance # plus (neg x); self#trace (Retrieval x); x) else zero method history = List.rev history end class type client_view = object method deposit : m -> unit method history : m operation list method withdraw : m -> m method balance : m end class virtual check_client x = let y = if (m 100.)#leq x then x else raise (Failure "Insufficient initial deposit") in object (self) initializer self#deposit y method virtual deposit: m -> unit end module Client (B : sig class bank : client_view end) = struct class account x : client_view = object inherit B.bank inherit check_client x end let discount x = let c = new account x in if today() < (1998,10,30) then c # deposit (m 100.); c end end;;

This shows the use of modules to group several class definitions that can in fact be thought of as a single unit. This unit would be provided by a bank for both internal and external uses. This is implemented as a functor that abstracts over the currency so that the same code can be used to provide accounts in different currencies.

The class bank is the real implementation of the bank account (it could have been inlined). This is the one that will be used for further extensions, refinements, etc. Conversely, the client will only be given the client view.

# module Euro_account = Account(Euro);;
# module Client = Euro_account.Client (Euro_account);;
# new Client.account (new Euro.c 100.);;

Hence, the clients do not have direct access to the balance, nor the history of their own accounts. Their only way to change their balance is to deposit or withdraw money. It is important to give the clients a class and not just the ability to create accounts (such as the promotional discount account), so that they can personalize their account. For instance, a client may refine the deposit and withdraw methods so as to do his own financial bookkeeping, automatically. On the other hand, the function discount is given as such, with no possibility for further personalization.

It is important to provide the client’s view as a functor Client so that client accounts can still be built after a possible specialization of the bank. The functor Client may remain unchanged and be passed the new definition to initialize a client’s view of the extended account.

# module Investment_account (M : MONEY) = struct type m = M.c module A = Account(M) class bank = object inherit A.bank as super method deposit x = if (new M.c 1000.)#leq x then print_string "Would you like to invest?"; super#deposit x end module Client = A.Client end;;

The functor Client may also be redefined when some new features of the account can be given to the client.

# module Internet_account (M : MONEY) = struct type m = M.c module A = Account(M) class bank = object inherit A.bank method mail s = print_string s end class type client_view = object method deposit : m -> unit method history : m operation list method withdraw : m -> m method balance : m method mail : string -> unit end module Client (B : sig class bank : client_view end) = struct class account x : client_view = object inherit B.bank inherit A.check_client x end end end;;

8.2 Simple modules as classes

One may wonder whether it is possible to treat primitive types such as integers and strings as objects. Although this is usually uninteresting for integers or strings, there may be some situations where this is desirable. The class money above is such an example. We show here how to do it for strings.

8.2.1 Strings

A naive definition of strings as objects could be:

# class ostring s = object method get n = String.get s n method print = print_string s method escaped = new ostring (String.escaped s) end;;
class ostring : string -> object method escaped : ostring method get : int -> char method print : unit end

However, the method escaped returns an object of the class ostring, and not an object of the current class. Hence, if the class is further extended, the method escaped will only return an object of the parent class.

# class sub_string s = object inherit ostring s method sub start len = new sub_string (String.sub s start len) end;;
class sub_string : string -> object method escaped : ostring method get : int -> char method print : unit method sub : int -> int -> sub_string end

As seen in section 3.16, the solution is to use functional update instead. We need to create an instance variable containing the representation s of the string.

# class better_string s = object val repr = s method get n = String.get repr n method print = print_string repr method escaped = {< repr = String.escaped repr >} method sub start len = {< repr = String.sub s start len >} end;;
class better_string : string -> object ('a) val repr : string method escaped : 'a method get : int -> char method print : unit method sub : int -> int -> 'a end

As shown in the inferred type, the methods escaped and sub now return objects of the same type as the one of the class.

Another difficulty is the implementation of the method concat. In order to concatenate a string with another string of the same class, one must be able to access the instance variable externally. Thus, a method repr returning s must be defined. Here is the correct definition of strings:

# class ostring s = object (self : 'mytype) val repr = s method repr = repr method get n = String.get repr n method print = print_string repr method escaped = {< repr = String.escaped repr >} method sub start len = {< repr = String.sub s start len >} method concat (t : 'mytype) = {< repr = repr ^ t#repr >} end;;
class ostring : string -> object ('a) val repr : string method concat : 'a -> 'a method escaped : 'a method get : int -> char method print : unit method repr : string method sub : int -> int -> 'a end

Another constructor of the class string can be defined to return a new string of a given length:

# class cstring n = ostring (String.make n ' ');;
class cstring : int -> ostring

Here, exposing the representation of strings is probably harmless. We do could also hide the representation of strings as we hid the currency in the class money of section 3.17.

Stacks

There is sometimes an alternative between using modules or classes for parametric data types. Indeed, there are situations when the two approaches are quite similar. For instance, a stack can be straightforwardly implemented as a class:

# exception Empty;;
exception Empty
# class ['a] stack = object val mutable l = ([] : 'a list) method push x = l <- x::l method pop = match l with [] -> raise Empty | a::l' -> l <- l'; a method clear = l <- [] method length = List.length l end;;
class ['a] stack : object val mutable l : 'a list method clear : unit method length : int method pop : 'a method push : 'a -> unit end

However, writing a method for iterating over a stack is more problematic. A method fold would have type ('b -> 'a -> 'b) -> 'b -> 'b. Here 'a is the parameter of the stack. The parameter 'b is not related to the class 'a stack but to the argument that will be passed to the method fold. A naive approach is to make 'b an extra parameter of class stack:

# class ['a, 'b] stack2 = object inherit ['a] stack method fold f (x : 'b) = List.fold_left f x l end;;
class ['a, 'b] stack2 : object val mutable l : 'a list method clear : unit method fold : ('b -> 'a -> 'b) -> 'b -> 'b method length : int method pop : 'a method push : 'a -> unit end

However, the method fold of a given object can only be applied to functions that all have the same type:

# let s = new stack2;;
val s : ('_weak1, '_weak2) stack2 = <obj>
# s#fold ( + ) 0;;
- : int = 0
# s;;
- : (int, int) stack2 = <obj>

A better solution is to use polymorphic methods, which were introduced in OCaml version 3.05. Polymorphic methods makes it possible to treat the type variable 'b in the type of fold as universally quantified, giving fold the polymorphic type Forall 'b. ('b -> 'a -> 'b) -> 'b -> 'b. An explicit type declaration on the method fold is required, since the type checker cannot infer the polymorphic type by itself.

# class ['a] stack3 = object inherit ['a] stack method fold : 'b. ('b -> 'a -> 'b) -> 'b -> 'b = fun f x -> List.fold_left f x l end;;
class ['a] stack3 : object val mutable l : 'a list method clear : unit method fold : ('b -> 'a -> 'b) -> 'b -> 'b method length : int method pop : 'a method push : 'a -> unit end

8.2.2 Hashtbl

A simplified version of object-oriented hash tables should have the following class type.

# class type ['a, 'b] hash_table = object method find : 'a -> 'b method add : 'a -> 'b -> unit end;;
class type ['a, 'b] hash_table = object method add : 'a -> 'b -> unit method find : 'a -> 'b end

A simple implementation, which is quite reasonable for small hash tables is to use an association list:

# class ['a, 'b] small_hashtbl : ['a, 'b] hash_table = object val mutable table = [] method find key = List.assoc key table method add key value = table <- (key, value) :: table end;;
class ['a, 'b] small_hashtbl : ['a, 'b] hash_table

A better implementation, and one that scales up better, is to use a true hash table
 whose elements are small hash tables!

# class ['a, 'b] hashtbl size : ['a, 'b] hash_table = object (self) val table = Array.init size (fun i -> new small_hashtbl) method private hash key = (Hashtbl.hash key) mod (Array.length table) method find key = table.(self#hash key) # find key method add key = table.(self#hash key) # add key end;;
class ['a, 'b] hashtbl : int -> ['a, 'b] hash_table

8.2.3 Sets

Implementing sets leads to another difficulty. Indeed, the method union needs to be able to access the internal representation of another object of the same class.

This is another instance of friend functions as seen in section 3.17. Indeed, this is the same mechanism used in the module Set in the absence of objects.

In the object-oriented version of sets, we only need to add an additional method tag to return the representation of a set. Since sets are parametric in the type of elements, the method tag has a parametric type 'a tag, concrete within the module definition but abstract in its signature. From outside, it will then be guaranteed that two objects with a method tag of the same type will share the same representation.

# module type SET = sig type 'a tag class ['a] c : object ('b) method is_empty : bool method mem : 'a -> bool method add : 'a -> 'b method union : 'b -> 'b method iter : ('a -> unit) -> unit method tag : 'a tag end end;;
# module Set : SET = struct let rec merge l1 l2 = match l1 with [] -> l2 | h1 :: t1 -> match l2 with [] -> l1 | h2 :: t2 -> if h1 < h2 then h1 :: merge t1 l2 else if h1 > h2 then h2 :: merge l1 t2 else merge t1 l2 type 'a tag = 'a list class ['a] c = object (_ : 'b) val repr = ([] : 'a list) method is_empty = (repr = []) method mem x = List.exists (( = ) x) repr method add x = {< repr = merge [x] repr >} method union (s : 'b) = {< repr = merge repr s#tag >} method iter (f : 'a -> unit) = List.iter f repr method tag = repr end end;;

8.3 The subject/observer pattern

The following example, known as the subject/observer pattern, is often presented in the literature as a difficult inheritance problem with inter-connected classes. The general pattern amounts to the definition a pair of two classes that recursively interact with one another.

The class observer has a distinguished method notify that requires two arguments, a subject and an event to execute an action.

# class virtual ['subject, 'event] observer = object method virtual notify : 'subject -> 'event -> unit end;;
class virtual ['subject, 'event] observer : object method virtual notify : 'subject -> 'event -> unit end

The class subject remembers a list of observers in an instance variable, and has a distinguished method notify_observers to broadcast the message notify to all observers with a particular event e.

# class ['observer, 'event] subject = object (self) val mutable observers = ([]:'observer list) method add_observer obs = observers <- (obs :: observers) method notify_observers (e : 'event) = List.iter (fun x -> x#notify self e) observers end;;
class ['a, 'event] subject : object ('b) constraint 'a = < notify : 'b -> 'event -> unit; .. > val mutable observers : 'a list method add_observer : 'a -> unit method notify_observers : 'event -> unit end

The difficulty usually lies in defining instances of the pattern above by inheritance. This can be done in a natural and obvious manner in OCaml, as shown on the following example manipulating windows.

# type event = Raise | Resize | Move;;
type event = Raise | Resize | Move
# let string_of_event = function Raise -> "Raise" | Resize -> "Resize" | Move -> "Move";;
val string_of_event : event -> string = <fun>
# let count = ref 0;;
val count : int ref = {contents = 0}
# class ['observer] window_subject = let id = count := succ !count; !count in object (self) inherit ['observer, event] subject val mutable position = 0 method identity = id method move x = position <- position + x; self#notify_observers Move method draw = Printf.printf "{Position = %d}\n" position; end;;
class ['a] window_subject : object ('b) constraint 'a = < notify : 'b -> event -> unit; .. > val mutable observers : 'a list val mutable position : int method add_observer : 'a -> unit method draw : unit method identity : int method move : int -> unit method notify_observers : event -> unit end
# class ['subject] window_observer = object inherit ['subject, event] observer method notify s e = s#draw end;;
class ['a] window_observer : object constraint 'a = < draw : unit; .. > method notify : 'a -> event -> unit end

As can be expected, the type of window is recursive.

# let window = new window_subject;;
val window : (< notify : 'a -> event -> unit; .. > as '_weak3) window_subject as 'a = <obj>

However, the two classes of window_subject and window_observer are not mutually recursive.

# let window_observer = new window_observer;;
val window_observer : (< draw : unit; .. > as '_weak4) window_observer = <obj>
# window#add_observer window_observer;;
- : unit = ()
# window#move 1;;
{Position = 1} - : unit = ()

Classes window_observer and window_subject can still be extended by inheritance. For instance, one may enrich the subject with new behaviors and refine the behavior of the observer.

# class ['observer] richer_window_subject = object (self) inherit ['observer] window_subject val mutable size = 1 method resize x = size <- size + x; self#notify_observers Resize val mutable top = false method raise = top <- true; self#notify_observers Raise method draw = Printf.printf "{Position = %d; Size = %d}\n" position size; end;;
class ['a] richer_window_subject : object ('b) constraint 'a = < notify : 'b -> event -> unit; .. > val mutable observers : 'a list val mutable position : int val mutable size : int val mutable top : bool method add_observer : 'a -> unit method draw : unit method identity : int method move : int -> unit method notify_observers : event -> unit method raise : unit method resize : int -> unit end
# class ['subject] richer_window_observer = object inherit ['subject] window_observer as super method notify s e = if e <> Raise then s#raise; super#notify s e end;;
class ['a] richer_window_observer : object constraint 'a = < draw : unit; raise : unit; .. > method notify : 'a -> event -> unit end

We can also create a different kind of observer:

# class ['subject] trace_observer = object inherit ['subject, event] observer method notify s e = Printf.printf "<Window %d <== %s>\n" s#identity (string_of_event e) end;;
class ['a] trace_observer : object constraint 'a = < identity : int; .. > method notify : 'a -> event -> unit end

and attach several observers to the same object:

# let window = new richer_window_subject;;
val window : (< notify : 'a -> event -> unit; .. > as '_weak5) richer_window_subject as 'a = <obj>
# window#add_observer (new richer_window_observer);;
- : unit = ()
# window#add_observer (new trace_observer);;
- : unit = ()
# window#move 1; window#resize 2;;
<Window 1 <== Move> <Window 1 <== Raise> {Position = 1; Size = 1} {Position = 1; Size = 1} <Window 1 <== Resize> <Window 1 <== Raise> {Position = 1; Size = 3} {Position = 1; Size = 3} - : unit = ()

Chapter 9 Parallel programming

In this chapter, we shall look at the parallel programming facilities in OCaml. The OCaml standard library exposes low-level primitives for parallel programming. We recommend the users to utilise higher-level parallel programming libraries such as domainslib. This tutorial will first cover the high-level parallel programming using domainslib followed by low-level primitives exposed by the compiler.

OCaml distinguishes concurrency and parallelism and provides distinct mechanisms for expressing them. Concurrency is overlapped execution of tasks (section 12.24.2) whereas parallelism is simultaneous execution of tasks. In particular, parallel tasks overlap in time but concurrent tasks may or may not overlap in time. Tasks may execute concurrently by yielding control to each other. While concurrency is a program structuring mechanism, parallelism is a mechanism to make your programs run faster. If you are interested in the concurrent programming mechanisms in OCaml, please refer to the section 12.24 on effect handlers and the chapter 33 on the threads library.

9.1 Domains

Domains are the units of parallelism in OCaml. The module Domain provides the primitives to create and manage domains. New domains can be spawned using the spawn function.

Domain.spawn (fun _ -> print_endline "I ran in parallel")
I ran in parallel - : unit Domain.t = <abstr>

The spawn function executes the given computation in parallel with the calling domain.

Domains are heavy-weight entities. Each domain maps 1:1 to an operating system thread. Each domain also has its own runtime state, which includes domain-local structures for allocating memory. Hence, they are relatively expensive to create and tear down.

It is recommended that the programs do not spawn more domains than cores available.

In this tutorial, we shall be implementing, running and measuring the performance of parallel programs. The results observed are dependent on the number of cores available on the target machine. This tutorial is being written on a 2.3 GHz Quad-Core Intel Core i7 MacBook Pro with 4 cores and 8 hardware threads. It is reasonable to expect roughly 4x performance on 4 domains for parallel programs with little coordination between the domains, and when the machine is not under load. Beyond 4 domains, the speedup is likely to be less than linear. We shall also use the command-line benchmarking tool hyperfine for benchmarking our programs.

9.1.1 Joining domains

We shall use the program to compute the nth Fibonacci number using recursion as a running example. The sequential program for computing the nth Fibonacci number is given below.

(* fib.ml *) let n = try int_of_string Sys.argv.(1) with _ -> 1 let rec fib n = if n < 2 then 1 else fib (n - 1) + fib (n - 2) let main () = let r = fib n in Printf.printf "fib(%d) = %d\n%!" n r let _ = main ()

The program can be compiled and benchmarked as follows.

$ ocamlopt -o fib.exe fib.ml
$ ./fib.exe 42
fib(42) = 433494437
$ hyperfine './fib.exe 42' # Benchmarking
Benchmark 1: ./fib.exe 42
  Time (mean ± sd):     1.193 s ±  0.006 s    [User: 1.186 s, System: 0.003 s]
  Range (min 
 max):    1.181 s 
  1.202 s    10 runs

We see that it takes around 1.2 seconds to compute the 42nd Fibonacci number.

Spawned domains can be joined using the join function to get their results. The join function waits for target domain to terminate. The following program computes the nth Fibonacci number twice in parallel.

(* fib_twice.ml *) let n = int_of_string Sys.argv.(1) let rec fib n = if n < 2 then 1 else fib (n - 1) + fib (n - 2) let main () = let d1 = Domain.spawn (fun _ -> fib n) in let d2 = Domain.spawn (fun _ -> fib n) in let r1 = Domain.join d1 in Printf.printf "fib(%d) = %d\n%!" n r1; let r2 = Domain.join d2 in Printf.printf "fib(%d) = %d\n%!" n r2 let _ = main ()

The program spawns two domains which compute the nth Fibonacci number. The spawn function returns a Domain.t value which can be joined to get the result of the parallel computation. The join function blocks until the computation runs to completion.

$ ocamlopt -o fib_twice.exe fib_twice.ml
$ ./fib_twice.exe 42
fib(42) = 433494437
fib(42) = 433494437
$ hyperfine './fib_twice.exe 42'
Benchmark 1: ./fib_twice.exe 42
  Time (mean ± sd):     1.249 s ±  0.025 s    [User: 2.451 s, System: 0.012 s]
  Range (min 
 max):    1.221 s 
  1.290 s    10 runs

As one can see that computing the nth Fibonacci number twice almost took the same time as computing it once thanks to parallelism.

9.2 Domainslib: A library for nested-parallel programming

Let us attempt to parallelise the Fibonacci function. The two recursive calls may be executed in parallel. However, naively parallelising the recursive calls by spawning domains for each one will not work as it spawns too many domains.

(* fib_par1.ml *) let n = try int_of_string Sys.argv.(1) with _ -> 1 let rec fib n = if n < 2 then 1 else begin let d1 = Domain.spawn (fun _ -> fib (n - 1)) in let d2 = Domain.spawn (fun _ -> fib (n - 2)) in Domain.join d1 + Domain.join d2 end let main () = let r = fib n in Printf.printf "fib(%d) = %d\n%!" n r let _ = main ()
fib(1) = 1 val n : int = 1 val fib : int -> int = <fun> val main : unit -> unit = <fun>
$ ocamlopt -o fib_par1.exe fib_par1.ml
$ ./fib_par1.exe 42
Fatal error: exception Failure("failed to allocate domain")

OCaml has a limit of 128 domains that can be active at the same time. An attempt to spawn more domains will raise an exception. How then can we parallelise the Fibonacci function?

9.2.1 Parallelising Fibonacci using domainslib

The OCaml standard library provides only low-level primitives for concurrent and parallel programming, leaving high-level programming libraries to be developed and distributed outside the core compiler distribution. Domainslib is such a library for nested-parallel programming, which is epitomised by the parallelism available in the recursive Fibonacci computation. Let us use domainslib to parallelise the recursive Fibonacci program. It is recommended that you install domainslib using the opam package manager. This tutorial uses domainslib version 0.5.0.

Domainslib provides an async/await mechanism for spawning parallel tasks and awaiting their results. On top of this mechanism, domainslib provides parallel iterators. At its core, domainslib has an efficient implementation of work-stealing queue in order to efficiently share tasks with other domains. A parallel implementation of the Fibonacci program is given below.

(* fib_par2.ml *)
let num_domains = int_of_string Sys.argv.(1)
let n = int_of_string Sys.argv.(2)

let rec fib n = if n < 2 then 1 else fib (n - 1) + fib (n - 2)

module T = Domainslib.Task

let rec fib_par pool n =
  if n > 20 then begin
    let a = T.async pool (fun _ -> fib_par pool (n-1)) in
    let b = T.async pool (fun _ -> fib_par pool (n-2)) in
    T.await pool a + T.await pool b
  end else fib n

let main () =
  let pool = T.setup_pool ~num_domains:(num_domains - 1) () in
  let res = T.run pool (fun _ -> fib_par pool n) in
  T.teardown_pool pool;
  Printf.printf "fib(%d) = %d\n" n res

let _ = main ()

The program takes the number of domains and the input to the Fibonacci function as the first and the second command-line arguments respectively.

Let us start with the main function. First, we set up a pool of domains on which the nested parallel tasks will run. The domain invoking the run function will also participate in executing the tasks submitted to the pool. We invoke the parallel Fibonacci function fib_par in the run function. Finally, we tear down the pool and print the result.

For sufficiently large inputs (n > 20), the fib_par function spawns the left and the right recursive calls asynchronously in the pool using the async function. The async function returns a promise for the result. The result of an asynchronous computation is obtained by awaiting the promise using the await function. The await function call blocks until the promise is resolved.

For small inputs, the fib_par function simply calls the sequential Fibonacci function fib. It is important to switch to sequential mode for small problem sizes. If not, the cost of parallelisation will outweigh the work available.

For simplicity, we use ocamlfind to compile this program. It is recommended that the users use dune to build their programs that utilise libraries installed through opam.

$ ocamlfind ocamlopt -package domainslib -linkpkg -o fib_par2.exe fib_par2.ml
$ ./fib_par2.exe 1 42
fib(42) = 433494437
$ hyperfine './fib.exe 42' './fib_par2.exe 2 42' \
            './fib_par2.exe 4 42' './fib_par2.exe 8 42'
Benchmark 1: ./fib.exe 42
  Time (mean ± sd):     1.217 s ±  0.018 s    [User: 1.203 s, System: 0.004 s]
  Range (min 
 max):    1.202 s 
  1.261 s    10 runs

Benchmark 2: ./fib_par2.exe 2 42
  Time (mean ± sd):    628.2 ms ±   2.9 ms    [User: 1243.1 ms, System: 4.9 ms]
  Range (min 
 max):   625.7 ms 
 634.5 ms    10 runs

Benchmark 3: ./fib_par2.exe 4 42
  Time (mean ± sd):    337.6 ms ±  23.4 ms    [User: 1321.8 ms, System: 8.4 ms]
  Range (min 
 max):   318.5 ms 
 377.6 ms    10 runs

Benchmark 4: ./fib_par2.exe 8 42
  Time (mean ± sd):    250.0 ms ±   9.4 ms    [User: 1877.1 ms, System: 12.6 ms]
  Range (min 
 max):   242.5 ms 
 277.3 ms    11 runs

Summary
  './fib_par2.exe 8 42' ran
    1.35 ± 0.11 times faster than './fib_par2.exe 4 42'
    2.51 ± 0.10 times faster than './fib_par2.exe 2 42'
    4.87 ± 0.20 times faster than './fib.exe 42'

The results show that, with 8 domains, the parallel Fibonacci program runs 4.87 times faster than the sequential version.

9.2.2 Parallel iteration constructs

Many numerical algorithms use for-loops. The parallel-for primitive provides a straight-forward way to parallelise such code. Let us take the spectral-norm benchmark from the computer language benchmarks game and parallelise it. The sequential version of the program is given below.

(* spectralnorm.ml *) let n = try int_of_string Sys.argv.(1) with _ -> 32 let eval_A i j = 1. /. float((i+j)*(i+j+1)/2+i+1) let eval_A_times_u u v = let n = Array.length v - 1 in for i = 0 to n do let vi = ref 0. in for j = 0 to n do vi := !vi +. eval_A i j *. u.(j) done; v.(i) <- !vi done let eval_At_times_u u v = let n = Array.length v - 1 in for i = 0 to n do let vi = ref 0. in for j = 0 to n do vi := !vi +. eval_A j i *. u.(j) done; v.(i) <- !vi done let eval_AtA_times_u u v = let w = Array.make (Array.length u) 0.0 in eval_A_times_u u w; eval_At_times_u w v let () = let u = Array.make n 1.0 and v = Array.make n 0.0 in for _i = 0 to 9 do eval_AtA_times_u u v; eval_AtA_times_u v u done; let vv = ref 0.0 and vBv = ref 0.0 in for i=0 to n-1 do vv := !vv +. v.(i) *. v.(i); vBv := !vBv +. u.(i) *. v.(i) done; Printf.printf "%0.9f\n" (sqrt(!vBv /. !vv))

Observe that the program has nested loops in eval_A_times_u and eval_At_times_u. Each iteration of the outer loop body reads from u but writes to disjoint memory locations in v. Hence, the iterations of the outer loop are not dependent on each other and can be executed in parallel.

The parallel version of spectral norm is shown below.

(* spectralnorm_par.ml *)
let num_domains = try int_of_string Sys.argv.(1) with _ -> 1
let n = try int_of_string Sys.argv.(2) with _ -> 32

let eval_A i j = 1. /. float((i+j)*(i+j+1)/2+i+1)

module T = Domainslib.Task

let eval_A_times_u pool u v =
  let n = Array.length v - 1 in
  T.parallel_for pool ~start:0 ~finish:n ~body:(fun i ->
    let vi = ref 0. in
    for j = 0 to n do vi := !vi +. eval_A i j *. u.(j) done;
    v.(i) <- !vi
  )

let eval_At_times_u pool u v =
  let n = Array.length v - 1 in
  T.parallel_for pool ~start:0 ~finish:n ~body:(fun i ->
    let vi = ref 0. in
    for j = 0 to n do vi := !vi +. eval_A j i *. u.(j) done;
    v.(i) <- !vi
  )

let eval_AtA_times_u pool u v =
  let w = Array.make (Array.length u) 0.0 in
  eval_A_times_u pool u w; eval_At_times_u pool w v

let () =
  let pool = T.setup_pool ~num_domains:(num_domains - 1) () in
  let u = Array.make n 1.0  and  v = Array.make n 0.0 in
  T.run pool (fun _ ->
  for _i = 0 to 9 do
    eval_AtA_times_u pool u v; eval_AtA_times_u pool v u
  done);

  let vv = ref 0.0  and  vBv = ref 0.0 in
  for i=0 to n-1 do
    vv := !vv +. v.(i) *. v.(i);
    vBv := !vBv +. u.(i) *. v.(i)
  done;
  T.teardown_pool pool;
  Printf.printf "%0.9f\n" (sqrt(!vBv /. !vv))

Observe that the parallel_for function is isomorphic to the for-loop in the sequential version. No other change is required except for the boiler plate code to set up and tear down the pools.

$ ocamlopt -o spectralnorm.exe spectralnorm.ml
$ ocamlfind ocamlopt -package domainslib -linkpkg -o spectralnorm_par.exe \
  spectralnorm_par.ml
$ hyperfine './spectralnorm.exe 4096' './spectralnorm_par.exe 2 4096' \
            './spectralnorm_par.exe 4 4096' './spectralnorm_par.exe 8 4096'
Benchmark 1: ./spectralnorm.exe 4096
  Time (mean ± sd):     1.989 s ±  0.013 s    [User: 1.972 s, System: 0.007 s]
  Range (min 
 max):    1.975 s 
  2.018 s    10 runs

Benchmark 2: ./spectralnorm_par.exe 2 4096
  Time (mean ± sd):     1.083 s ±  0.015 s    [User: 2.140 s, System: 0.009 s]
  Range (min 
 max):    1.064 s 
  1.102 s    10 runs

Benchmark 3: ./spectralnorm_par.exe 4 4096
  Time (mean ± sd):    698.7 ms ±  10.3 ms    [User: 2730.8 ms, System: 18.3 ms]
  Range (min 
 max):   680.9 ms 
 721.7 ms    10 runs

Benchmark 4: ./spectralnorm_par.exe 8 4096
  Time (mean ± sd):    921.8 ms ±  52.1 ms    [User: 6711.6 ms, System: 51.0 ms]
  Range (min 
 max):   838.6 ms 
 989.2 ms    10 runs

Summary
  './spectralnorm_par.exe 4 4096' ran
    1.32 ± 0.08 times faster than './spectralnorm_par.exe 8 4096'
    1.55 ± 0.03 times faster than './spectralnorm_par.exe 2 4096'
    2.85 ± 0.05 times faster than './spectralnorm.exe 4096'

On the author’s machine, the program scales reasonably well up to 4 domains but performs worse with 8 domains. Recall that the machine only has 4 physical cores. Debugging and fixing this performance issue is beyond the scope of this tutorial.

9.3 Parallel garbage collection

An important aspect of the scalability of parallel OCaml programs is the scalability of the garbage collector (GC). The OCaml GC is designed to have both low latency and good parallel scalability. OCaml has a generational garbage collector with a small minor heap and a large major heap. New objects (upto a certain size) are allocated in the minor heap. Each domain has its own domain-local minor heap arena into which new objects are allocated without synchronising with the other domains. When a domain exhausts its minor heap arena, it calls for a stop-the-world collection of the minor heaps. In the stop-the-world section, all the domains collect their minor heap arenas in parallel evacuating the survivors to the major heap.

For the major heap, each domain maintains domain-local, size-segmented pools of memory into which large objects and survivors from the minor collection are allocated. Having domain-local pools avoids synchronisation for most major heap allocations. The major heap is collected by a concurrent mark-and-sweep algorithm that involves a few short stop-the-world pauses for each major cycle.

Overall, the users should expect the garbage collector to scale well with increasing number of domains, with the latency remaining low. For more information on the design and evaluation of the garbage collector, please have a look at the ICFP 2020 paper on Retrofitting Parallelism onto OCaml.

9.4 Memory model: The easy bits

Modern processors and compilers aggressively optimise programs. These optimisations accelerate without otherwise affecting sequential programs, but cause surprising behaviours to be visible in parallel programs. To benefit from these optimisations, OCaml adopts a relaxed memory model that precisely specifies which of these relaxed behaviours programs may observe. While these models are difficult to program against directly, the OCaml memory model provides recipes that retain the simplicity of sequential reasoning.

Firstly, immutable values may be freely shared between multiple domains and may be accessed in parallel. For mutable data structures such as reference cells, arrays and mutable record fields, programmers should avoid data races. Reference cells, arrays and mutable record fields are said to be non-atomic data structures. A data race is said to occur when two domains concurrently access a non-atomic memory location without synchronisation and at least one of the accesses is a write. OCaml provides a number of ways to introduce synchronisation including atomic variables (section 9.7) and mutexes (section 9.5).

Importantly, for data race free (DRF) programs, OCaml provides sequentially consistent (SC) semantics – the observed behaviour of such programs can be explained by the interleaving of operations from different domains. This property is known as DRF-SC guarantee. Moreover, in OCaml, DRF-SC guarantee is modular – if a part of a program is data race free, then the OCaml memory model ensures that those parts have sequential consistency despite other parts of the program having data races. Even for programs with data races, OCaml provides strong guarantees. While the user may observe non sequentially consistent behaviours, there are no crashes.

For more details on the relaxed behaviours in the presence of data races, please have a look at the chapter on the hard bits of the memory model (chapter 10).

9.5 Blocking synchronisation

Domains may perform blocking synchronisation with the help of Mutex, Condition and Semaphore modules. These modules are the same as those used to synchronise threads created by the threads library (chapter 33). For clarity, in the rest of this chapter, we shall call the threads created by the threads library as systhreads. The following program implements a concurrent stack using mutex and condition variables.

module Blocking_stack : sig type 'a t val make : unit -> 'a t val push : 'a t -> 'a -> unit val pop : 'a t -> 'a end = struct type 'a t = { mutable contents: 'a list; mutex : Mutex.t; nonempty : Condition.t } let make () = { contents = []; mutex = Mutex.create (); nonempty = Condition.create () } let push r v = Mutex.lock r.mutex; r.contents <- v::r.contents; Condition.signal r.nonempty; Mutex.unlock r.mutex let pop r = Mutex.lock r.mutex; let rec loop () = match r.contents with | [] -> Condition.wait r.nonempty r.mutex; loop () | x::xs -> r.contents <- xs; x in let res = loop () in Mutex.unlock r.mutex; res end

The concurrent stack is implemented using a record with three fields: a mutable field contents which stores the elements in the stack, a mutex to control access to the contents field, and a condition variable nonempty, which is used to signal blocked domains waiting for the stack to become non-empty.

The push operation locks the mutex, updates the contents field with a new list whose head is the element being pushed and the tail is the old list. The condition variable nonempty is signalled while the lock is held in order to wake up any domains waiting on this condition. If there are waiting domains, one of the domains is woken up. If there are none, then the signal operation has no effect.

The pop operation locks the mutex and checks whether the stack is empty. If so, the calling domain waits on the condition variable nonempty using the wait primitive. The wait call atomically suspends the execution of the current domain and unlocks the mutex. When this domain is woken up again (when the wait call returns), it holds the lock on mutex. The domain tries to read the contents of the stack again. If the pop operation sees that the stack is non-empty, it updates the contents to the tail of the old list, and returns the head.

The use of mutex to control access to the shared resource contents introduces sufficient synchronisation between multiple domains using the stack. Hence, there are no data races when multiple domains use the stack in parallel.

9.5.1 Interaction with systhreads

How do systhreads interact with domains? The systhreads created on a particular domain remain pinned to that domain. Only one systhread at a time is allowed to run OCaml code on a particular domain. However, systhreads belonging to a particular domain may run C library or system code in parallel. Systhreads belonging to different domains may execute in parallel.

When using systhreads, the thread created for executing the computation given to Domain.spawn is also treated as a systhread. For example, the following program creates in total two domains (including the initial domain) with two systhreads each (including the initial systhread for each of the domains).

(* dom_thr.ml *)
let m = Mutex.create ()
let r = ref None (* protected by m *)

let task () =
  let my_thr_id = Thread.(id (self ())) in
  let my_dom_id :> int = Domain.self () in
  Mutex.lock m;
  begin match !r with
  | None ->
      Printf.printf "Thread %d running on domain %d saw initial write\n%!"
        my_thr_id my_dom_id
  | Some their_thr_id ->
      Printf.printf "Thread %d running on domain %d saw the write by thread %d\n%!"
        my_thr_id my_dom_id their_thr_id;
  end;
  r := Some my_thr_id;
  Mutex.unlock m

let task' () =
  let t = Thread.create task () in
  task ();
  Thread.join t

let main () =
  let d = Domain.spawn task' in
  task' ();
  Domain.join d

let _ = main ()
$ ocamlopt -I +threads unix.cmxa threads.cmxa -o dom_thr.exe dom_thr.ml
$ ./dom_thr.exe
Thread 1 running on domain 1 saw initial write
Thread 0 running on domain 0 saw the write by thread 1
Thread 2 running on domain 1 saw the write by thread 0
Thread 3 running on domain 0 saw the write by thread 2

This program uses a shared reference cell protected by a mutex to communicate between the different systhreads running on two different domains. The systhread identifiers uniquely identify systhreads in the program. The initial domain gets the domain id and the thread id as 0. The newly spawned domain gets domain id as 1.

9.6 Interaction with C bindings

During parallel execution with multiple domains, C code running on a domain may run in parallel with any C code running in other domains even if neither of them has released the “domain lock”. Prior to OCaml 5.0, C bindings may have assumed that if the OCaml runtime lock is not released, then it would be safe to manipulate global C state (e.g. initialise a function-local static value). This is no longer true in the presence of parallel execution with multiple domains.

9.7 Atomics

Mutex, condition variables and semaphores are used to implement blocking synchronisation between domains. For non-blocking synchronisation, OCaml provides Atomic variables. As the name suggests, non-blocking synchronisation does not provide mechanisms for suspending and waking up domains. On the other hand, primitives used in non-blocking synchronisation are often compiled to atomic read-modify-write primitives that the hardware provides. As an example, the following program increments a non-atomic counter and an atomic counter in parallel.

(* incr.ml *) let twice_in_parallel f = let d1 = Domain.spawn f in let d2 = Domain.spawn f in Domain.join d1; Domain.join d2 let plain_ref n = let r = ref 0 in let f () = for _i=1 to n do incr r done in twice_in_parallel f; Printf.printf "Non-atomic ref count: %d\n" !r let atomic_ref n = let r = Atomic.make 0 in let f () = for _i=1 to n do Atomic.incr r done in twice_in_parallel f; Printf.printf "Atomic ref count: %d\n" (Atomic.get r) let main () = let n = try int_of_string Sys.argv.(1) with _ -> 1 in plain_ref n; atomic_ref n let _ = main ()
$ ocamlopt -o incr.exe incr.ml
$ ./incr.exe 1_000_000
Non-atomic ref count: 1187193
Atomic ref count: 2000000

Observe that the result from using the non-atomic counter is lower than what one would naively expect. This is because the non-atomic incr function is equivalent to:

let incr r = let curr = !r in r := curr + 1

Observe that the load and the store are two separate operations, and the increment operation as a whole is not performed atomically. When two domains execute this code in parallel, both of them may read the same value of the counter curr and update it to curr + 1. Hence, instead of two increments, the effect will be that of a single increment. On the other hand, the atomic counter performs the load and the store atomically with the help of hardware support for atomicity. The atomic counter returns the expected result.

The atomic variables can be used for low-level synchronisation between the domains. The following example uses an atomic variable to exchange a message between two domains.

let r = Atomic.make None let sender () = Atomic.set r (Some "Hello") let rec receiver () = match Atomic.get r with | None -> Domain.cpu_relax (); receiver () | Some m -> print_endline m let main () = let s = Domain.spawn sender in let d = Domain.spawn receiver in Domain.join s; Domain.join d let _ = main ()
Hello val r : string option Atomic.t = <abstr> val sender : unit -> unit = <fun> val receiver : unit -> unit = <fun> val main : unit -> unit = <fun>

While the sender and the receiver compete to access r, this is not a data race since r is an atomic reference.

9.7.1 Lock-free stack

The Atomic module is used to implement non-blocking, lock-free data structures. The following program implements a lock-free stack.

module Lockfree_stack : sig type 'a t val make : unit -> 'a t val push : 'a t -> 'a -> unit val pop : 'a t -> 'a option end = struct type 'a t = 'a list Atomic.t let make () = Atomic.make [] let rec push r v = let s = Atomic.get r in if Atomic.compare_and_set r s (v::s) then () else (Domain.cpu_relax (); push r v) let rec pop r = let s = Atomic.get r in match s with | [] -> None | x::xs -> if Atomic.compare_and_set r s xs then Some x else (Domain.cpu_relax (); pop r) end

The atomic stack is represented by an atomic reference that holds a list. The push and pop operations use the compare_and_set primitive to attempt to atomically update the atomic reference. The expression compare_and_set r seen v sets the value of r to v if and only if its current value is physically equal to seen. Importantly, the comparison and the update occur atomically. The expression evaluates to true if the comparison succeeded (and the update happened) and false otherwise.

If the compare_and_set fails, then some other domain is also attempting to update the atomic reference at the same time. In this case, the push and pop operations call Domain.cpu_relax to back off for a short duration allowing competing domains to make progress before retrying the failed operation. This lock-free stack implementation is also known as Treiber stack.

Chapter 10 Memory model: The hard bits

This chapter describes the details of OCaml relaxed memory model. The relaxed memory model describes what values an OCaml program is allowed to witness when reading a memory location. If you are interested in high-level parallel programming in OCaml, please have a look at the parallel programming chapter 9.

This chapter is aimed at experts who would like to understand the details of the OCaml memory model from a practitioner’s perspective. For a formal definition of the OCaml memory model, its guarantees and the compilation to hardware memory models, please have a look at the PLDI 2018 paper on Bounding Data Races in Space and Time. The memory model presented in this chapter is an extension of the one presented in the PLDI 2018 paper. This chapter also covers some pragmatic aspects of the memory model that are not covered in the paper.

10.1 Why weakly consistent memory?

The simplest memory model that we could give to our programs is sequential consistency. Under sequential consistency, the values observed by the program can be explained through some interleaving of the operations from different domains in the program. For example, consider the following program with two domains d1 and d2 executing in parallel:

let d1 a b = let r1 = !a * 2 in let r2 = !b in let r3 = !a * 2 in (r1, r2, r3) let d2 b = b := 0 let main () = let a = ref 1 in let b = ref 1 in let h = Domain.spawn (fun _ -> let r1, r2, r3 = d1 a b in Printf.printf "r1 = %d, r2 = %d, r3 = %d\n" r1 r2 r3) in d2 b; Domain.join h

The reference cells a and b are initially 1. The user may observe r1 = 2, r2 = 0, r3 = 2 if the write to b in d2 occurred before the read of b in d1. Here, the observed behaviour can be explained in terms of interleaving of the operations from different domains.

Let us now assume that a and b are aliases of each other.

let d1 a b = let r1 = !a * 2 in let r2 = !b in let r3 = !a * 2 in (r1, r2, r3) let d2 b = b := 0 let main () = let ab = ref 1 in let h = Domain.spawn (fun _ -> let r1, r2, r3 = d1 ab ab in assert (not (r1 = 2 && r2 = 0 && r3 = 2))) in d2 ab; Domain.join h

In the above program, the variables ab, a and b refer to the same reference cell. One would expect that the assertion in the main function will never fail. The reasoning is that if r2 is 0, then the write in d2 occurred before the read of b in d1. Given that a and b are aliases, the second read of a in d1 should also return 0.

10.1.1 Compiler optimisations

Surprisingly, this assertion may fail in OCaml due to compiler optimisations. The OCaml compiler observes the common sub-expression !a * 2 in d1 and optimises the program to:

let d1 a b = let r1 = !a * 2 in let r2 = !b in let r3 = r1 in (* CSE: !a * 2 ==> r1 *) (r1, r2, r3) let d2 b = b := 0 let main () = let ab = ref 1 in let h = Domain.spawn (fun _ -> let r1, r2, r3 = d1 ab ab in assert (not (r1 = 2 && r2 = 0 && r3 = 2))) in d2 ab; Domain.join h

This optimisation is known as the common sub-expression elimination (CSE). Such optimisations are valid and necessary for good performance, and do not change the sequential meaning of the program. However, CSE breaks sequential reasoning.

In the optimized program above, even if the write to b in d2 occurs between the first and the second reads in d1, the program will observe the value 2 for r3, causing the assertion to fail. The observed behaviour cannot be explained by interleaving of operations from different domains in the source program. Thus, CSE optimization is said to be invalid under sequential consistency.

One way to explain the observed behaviour is as if the operations performed on a domain were reordered. For example, if the second and the third reads from d2 were reordered,

let d1 a b = let r1 = !a * 2 in let r3 = !a * 2 in let r2 = !b in (r1, r2, r3)

then we can explain the observed behaviour (2,0,2) returned by d1.

10.1.2 Hardware optimisations

The other source of reordering is by the hardware. Modern hardware architectures have complex cache hierarchies with multiple levels of cache. While cache coherence ensures that reads and writes to a single memory location respect sequential consistency, the guarantees on programs that operate on different memory locations are much weaker. Consider the following program:

let a = ref 0 and b = ref 0 let d1 () = a := 1; !b let d2 () = b := 1; !a let main () = let h = Domain.spawn d2 in let r1 = d1 () in let r2 = Domain.join h in assert (not (r1 = 0 && r2 = 0))

Under sequential consistency, we would never expect the assertion to fail. However, even on x86, which offers much stronger guarantees than ARM, the writes performed at a CPU core are not immediately published to all of the other cores. Since a and b are different memory locations, the reads of a and b may both witness the initial values, leading to the assertion failure.

This behaviour can be explained if a load is allowed to be reordered before a preceding store to a different memory location. This reordering can happen due to the presence of in-core store-buffers on modern processors. Each core effectively has a FIFO buffer of pending writes to avoid the need to block while a write completes. The writes to a and b may be in the store-buffers of cores c1 and c2 running the domains d1 and d2, respectively. The reads of b and a running on the cores c1 and c2, respectively, will not see the writes if the writes have not propagated from the buffers to the main memory.

10.2 Data race freedom implies sequential consistency

The aim of the OCaml relaxed memory model is to precisely describe which orders are preserved by the OCaml program. The compiler and the hardware are free to optimize the program as long as they respect the ordering guarantees of the memory model. While programming directly under the relaxed memory model is difficult, the memory model also describes the conditions under which a program will only exhibit sequentially consistent behaviours. This guarantee is known as data race freedom implies sequential consistency (DRF-SC). In this section, we shall describe this guarantee. In order to do this, we first need a number of definitions.

10.2.1 Memory locations

OCaml classifies memory locations into atomic and non-atomic locations. Reference cells, array fields and mutable record fields are non-atomic memory locations. Immutable objects are non-atomic locations with an initialising write but no further updates. Atomic memory locations are those that are created using the Atomic module.

10.2.2 Happens-before relation

Let us imagine that the OCaml programs are executed by an abstract machine that executes one action at a time, arbitrarily picking one of the available domains at each step. We classify actions into two: inter-domain and intra-domain. An inter-domain action is one which can be observed and be influenced by actions on other domains. There are several inter-domain actions:

On the other hand, intra-domain actions can neither be observed nor influence the execution of other domains. Examples include evaluating an arithmetic expression, calling a function, etc. The memory model specification ignores such intra-domain actions. In the sequel, we use the term action to indicate inter-domain actions.

A totally ordered list of actions executed by the abstract machine is called an execution trace. There might be several possible execution traces for a given program due to non-determinism.

For a given execution trace, we define an irreflexive, transitive happens-before relation that captures the causality between actions in the OCaml program. The happens-before relation is defined as the smallest transitive relation satisfying the following properties:

10.2.3 Data race

In a given trace, two actions are said to be conflicting if they access the same non-atomic location, at least one is a write and neither is an initialising write to that location.

We say that a program has a data race if there exists some execution trace of the program with two conflicting actions and there does not exist a happens-before relationship between the conflicting accesses. A program without data races is said to be correctly synchronised.

10.2.4 DRF-SC

DRF-SC guarantee: A program without data races will only exhibit sequentially consistent behaviours.

DRF-SC is a strong guarantee for the programmers. Programmers can use sequential reasoning i.e., reasoning by executing one inter-domain action after the other, to identify whether their program has a data race. In particular, they do not need to reason about reorderings described in section 10.1 in order to determine whether their program has a data race. Once the determination that a particular program is data race free is made, they do not need to worry about reorderings in their code.

10.3 Reasoning with DRF-SC

In this section, we will look at examples of using DRF-SC for program reasoning. In this section, we will use the functions with names dN to represent domains executing in parallel with other domains. That is, we assume that there is a main function that runs the dN functions in parallel as follows:

let main () =
  let h1 = Domain.spawn d1 in
  let h2 = Domain.spawn d2 in
  ...
  ignore @@ Domain.join h1;
  ignore @@ Domain.join h2

Here is a simple example with a data race:

(* Has data race *) let r = ref 0 let d1 () = r := 1 let d2 () = !r

r is a non-atomic reference. The two domains race to access the reference, and d1 is a write. Since there is no happens-before relationship between the conflicting accesses, there is a data race.

Both of the programs that we had seen in the section 10.1 have data races. It is no surprise that they exhibit non sequentially consistent behaviours.

Accessing disjoint array indices and fields of a record in parallel is not a data race. For example,

(* No data race *) let a = [| 0; 1 |] let d1 () = a.(0) <- 42 let d2 () = a.(1) <- 42
(* No data race *) type t = { mutable a : int; mutable b : int } let r = {a = 0; b = 1} let d1 () = r.a <- 42 let d2 () = r.b <- 42

do not have data races.

Races on atomic locations do not lead to a data race.

(* No data race *) let r = Atomic.make 0 let d1 () = Atomic.set r 1 let d2 () = Atomic.get r

Message-passing

Atomic variables may be used for implementing non-blocking communication between domains.

(* No data race *) let msg = ref 0 let flag = Atomic.make false let d1 () = msg := 42; (* a *) Atomic.set flag true (* b *) let d2 () = if Atomic.get flag (* c *) then !msg (* d *) else 0

Observe that the actions a and d write and read from the same non-atomic location msg, respectively, and hence are conflicting. We need to establish that a and d have a happens-before relationship in order to show that this program does not have a data race.

The action a precedes b in program order, and hence, a happens-before b. Similarly, c happens-before d. If d2 observes the atomic variable flag to be true, then b precedes c in happens-before order. Since happens-before is transitive, the conflicting actions a and d are in happens-before order. If d2 observes the flag to be false, then the read of msg is not done. Hence, there is no conflicting access in this execution trace. Hence, the program does not have a data race.

The following modified version of the message passing program does have a data race.

(* Has data race *) let msg = ref 0 let flag = Atomic.make false let d1 () = msg := 42; (* a *) Atomic.set flag true (* b *) let d2 () = ignore (Atomic.get flag); (* c *) !msg (* d *)

The domain d2 now unconditionally reads the non-atomic reference msg. Consider the execution trace:

Atomic.get flag; (* c *)
!msg; (* d *)
msg := 42; (* a *)
Atomic.set flag true (* b *)

In this trace, d and a are conflicting operations. But there is no happens-before relationship between them. Hence, this program has a data race.

10.4 Local data race freedom

The OCaml memory model offers strong guarantees even for programs with data races. It offers what is known as local data race freedom sequential consistency (LDRF-SC) guarantee. A formal definition of this property is beyond the scope of this manual chapter. Interested readers are encouraged to read the PLDI 2018 paper on Bounding Data Races in Space and Time.

Informally, LDRF-SC says that the data race free parts of the program remain sequentially consistent. That is, even if the program has data races, those parts of the program that are disjoint from the parts with data races are amenable to sequential reasoning.

Consider the following snippet:

let snippet () = let c = ref 0 in c := 42; let a = !c in (a, c)

Observe that c is a newly allocated reference. Can the read of c return a value which is not 42? That is, can a ever be not 42? Surprisingly, in the C++ and Java memory models, the answer is yes. With the C++ memory model, if the program has a data race, even in unrelated parts, then the semantics is undefined. If this snippet were linked with a library that had a data race, then, under the C++ memory model, the read may return any value. Since data races on unrelated locations can affect program behaviour, we say that C++ memory model is not bounded in space.

Unlike C++, Java memory model is bounded in space. But Java memory model is not bounded in time; data races in the future will affect the past behaviour. For example, consider the translation of this example to Java. We assume a prior definition of Class c {int x;} and a shared non-volatile variable C g. Now the snippet may be part of a larger program with parallel threads:

(* Thread 1 *)
C c = new C();
c.x = 42;
a = c.x;
g = c;

(* Thread 2 *)
g.x = 7;

The read of c.x and the write of g in the first thread are done on separate memory locations. Hence, the Java memory model allows them to be reordered. As a result, the write in the second thread may occur before the read of c.x, and hence, c.x returns 7.

The OCaml equivalent of the Java code above is:

let g = ref None let snippet () = let c = ref 0 in c := 42; let a = !c in (a, c) let d1 () = let (a,c) = snippet () in g := Some c; a let d2 () = match !g with | None -> () | Some c -> c := 7

Observe that there is a data race on both g and c. Consider only the first three instructions in snippet:

let c = ref 0 in
c := 42;
let a = !c in
...

The OCaml memory model is bounded both in space and time. The only memory location here is c. Reasoning only about this snippet, there is neither the data race in space (the race on g) nor in time (the future race on c). Hence, the snippet will have sequentially consistent behaviour, and the value returned by !c will be 42.

The OCaml memory model guarantees that even for programs with data races, memory safety is preserved. While programs with data races may observe non-sequentially consistent behaviours, they will not crash.

10.5 An operational view of the memory model

In this section, we describe the semantics of the OCaml memory model. A formal definition of the operational view of the memory model is presented in section 3 of the PLDI 2018 paper on Bounding Data Races in Space and Time. This section presents an informal description of the memory model with the help of an example.

Given an OCaml program, which may possibly contain data races, the operational semantics tells you the values that may be observed by the read of a memory location. For simplicity, we restrict the intra-thread actions to just the accesses to atomic and non-atomic locations, ignoring domain spawn and join operations, and the operations on mutexes.

We describe the semantics of the OCaml memory model in a straightforward small-step operational manner. That is, the semantics is described by an abstract machine that executes one action at a time, arbitrarily picking one of the available domains at each step. This is similar to the abstract machine that we had used to describe the happens-before relationship in section 10.2.2.

10.5.1 Non-atomic locations

In the semantics, we model non-atomic locations as finite maps from timestamps t to values v. We take timestamps to be rational numbers. The timestamps are totally ordered but dense; there is a timestamp between any two others.

For example,

a: [t1 -> 1; t2 -> 2]
b: [t3 -> 3; t4 -> 4; t5 -> 5]
c: [t6 -> 5; t7 -> 6; t8 -> 7]

represents three non-atomic locations a, b and c and their histories. The location a has two writes at timestamps t1 and t2 with values 1 and 2, respectively. When we write a: [t1 -> 1; t2 -> 2], we assume that t1 < t2. We assume that the locations are initialised with a history that has a single entry at timestamp 0 that maps to the initial value.

10.5.2 Domains

Each domain is equipped with a frontier, which is a map from non-atomic locations to timestamps. Intuitively, each domain’s frontier records, for each non-atomic location, the latest write known to the thread. More recent writes may have occurred, but are not guaranteed to be visible.

For example,

d1: [a -> t1; b -> t3; c -> t7]
d2: [a -> t1; b -> t4; c -> t7]

represents two domains d1 and d2 and their frontiers.

10.5.3 Non-atomic accesses

Let us now define the semantics of non-atomic reads and writes. Suppose domain d1 performs the read of b. For non-atomic reads, the domains may read an arbitrary element of the history for that location, as long as it is not older than the timestamp in the domains’s frontier. In this case, since d1 frontier at b is at t3, the read may return the value 3, 4 or 5. A non-atomic read does not change the frontier of the current domain.

Suppose domain d2 writes the value 10 to c (c := 10). We pick a new timestamp t9 for this write such that it is later than d2’s frontier at c. Note a subtlety here: this new timestamp might not be later than everything else in the history, but merely later than any other write known to the writing domain. Hence, t9 may be inserted in c’s history either (a) between t7 and t8 or (b) after t8. Let us pick the former option for our discussion. Since the new write appears after all the writes known by the domain d2 to the location c, d2’s frontier at c is also updated. The new state of the abstract machine is:

(* Non-atomic locations *)
a: [t1 -> 1; t2 -> 2]
b: [t3 -> 3; t4 -> 4; t5 -> 5]
c: [t6 -> 5; t7 -> 6; t9 -> 10; t8 -> 7] (* new write at t9 *)

(* Domains *)
d1: [a -> t1; b -> t3; c -> t7]
d2: [a -> t1; b -> t4; c -> t9] (* frontier updated at c *)

10.5.4 Atomic accesses

Atomic locations carry not only values but also synchronization information. We model atomic locations as a pair of the value held by that location and a frontier. The frontier models the synchronization information, which is merged with the frontiers of threads that operate on the location. In this way, non-atomic writes made by one thread can become known to another by communicating via an atomic location.

For example,

(* Atomic locations *)
A: 10, [a -> t1; b -> t5; c -> t7]
B: 5,  [a -> t2; b -> t4; c -> t6]

shows two atomic variables A and B with values 10 and 5, respectively, and frontiers of their own. We use upper-case variable names to indicate atomic locations.

During atomic reads, the frontier of the location is merged into that of the domain performing the read. For example, suppose d1 reads B. The read returns 5, and d1’s frontier updated by merging it with B’s frontier, choosing the later timestamp for each location. The abstract machine state before the atomic read is:

(* Non-atomic locations *)
a: [t1 -> 1; t2 -> 2]
b: [t3 -> 3; t4 -> 4; t5 -> 5]
c: [t6 -> 5; t7 -> 6; t9 -> 10; t8 -> 7]

(* Domains *)
d1: [a -> t1; b -> t3; c -> t7]
d2: [a -> t1; b -> t4; c -> t9]

(* Atomic locations *)
A: 10, [a -> t1; b -> t5; c -> t7]
B: 5,  [a -> t2; b -> t4; c -> t6]

As a result of the atomic read, the abstract machine state is updated to:

(* Non-atomic locations *)
a: [t1 -> 1; t2 -> 2]
b: [t3 -> 3; t4 -> 4; t5 -> 5]
c: [t6 -> 5; t7 -> 6; t9 -> 10; t8 -> 7]

(* Domains *)
d1: [a -> t2; b -> t4; c -> t7] (* frontier updated at a and b *)
d2: [a -> t1; b -> t4; c -> t9]

(* Atomic locations *)
A: 10, [a -> t1; b -> t5; c -> t7]
B: 5,  [a -> t2; b -> t4; c -> t6]

During atomic writes, the value held by the atomic location is updated. The frontiers of both the writing domain and that of the location being written to are updated to the merge of the two frontiers. For example, if d2 writes 20 to A in the current machine state, the machine state is updated to:

(* Non-atomic locations *)
a: [t1 -> 1; t2 -> 2]
b: [t3 -> 3; t4 -> 4; t5 -> 5]
c: [t6 -> 5; t7 -> 6; t9 -> 10; t8 -> 7]

(* Domains *)
d1: [a -> t2; b -> t4; c -> t7]
d2: [a -> t1; b -> t5; c -> t9] (* frontier updated at b *)

(* Atomic locations *)
A: 20, [a -> t1; b -> t5; c -> t9] (* value updated. frontier updated at c. *)
B: 5,  [a -> t2; b -> t4; c -> t6]

10.5.5 Reasoning with the semantics

Let us revisit an example from earlier (section 10.1).

let a = ref 0 and b = ref 0 let d1 () = a := 1; !b let d2 () = b := 1; !a let main () = let h = Domain.spawn d2 in let r1 = d1 () in let r2 = Domain.join h in assert (not (r1 = 0 && r2 = 0))

This program has a data race on a and b, and hence, the program may exhibit non sequentially consistent behaviour. Let us use the semantics to show that the program may exhibit r1 = 0 && r2 = 0.

The initial state of the abstract machine is:

(* Non-atomic locations *)
a: [t0 -> 0]
b: [t1 -> 0]

(* Domains *)
d1: [a -> t0; b -> t1]
d2: [a -> t0; b -> t1]

There are several possible schedules for executing this program. Let us consider the following schedule:

1: a := 1 @ d1
2: b := 1 @ d2
3: !b     @ d1
4: !a     @ d2

After the first action a:=1 by d1, the machine state is:

(* Non-atomic locations *)
a: [t0 -> 0; t2 -> 1] (* new write at t2 *)
b: [t1 -> 0]

(* Domains *)
d1: [a -> t2; b -> t1] (* frontier updated at a *)
d2: [a -> t0; b -> t1]

After the second action b:=1 by d2, the machine state is:

(* Non-atomic locations *)
a: [t0 -> 0; t2 -> 1]
b: [t1 -> 0; t3 -> 1] (* new write at t3 *)

(* Domains *)
d1: [a -> t2; b -> t1]
d2: [a -> t0; b -> t3] (* frontier updated at b *)

Now, for the third action !b by d1, observe that d1’s frontier at b is at t1. Hence, the read may return either 0 or 1. Let us assume that it returns 0. The machine state is not updated by the non-atomic read.

Similarly, for the fourth action !a by d2, d2’s frontier at a is at t0. Hence, this read may also return either 0 or 1. Let us assume that it returns 0. Hence, the assertion in the original program, assert (not (r1 = 0 && r2 = 0)), will fail for this particular execution.

10.6 Non-compliant operations

There are certain operations which are not memory model compliant.

Part II
The OCaml language

Chapter 11 The OCaml language

Foreword

This document is intended as a reference manual for the OCaml language. It lists the language constructs, and gives their precise syntax and informal semantics. It is by no means a tutorial introduction to the language. A good working knowledge of OCaml is assumed.

No attempt has been made at mathematical rigor: words are employed with their intuitive meaning, without further definition. As a consequence, the typing rules have been left out, by lack of the mathematical framework required to express them, while they are definitely part of a full formal definition of the language.

Notations

The syntax of the language is given in BNF-like notation. Terminal symbols are set in typewriter font (like this). Non-terminal symbols are set in italic font (like that). Square brackets [
] denote optional components. Curly brackets {
} denotes zero, one or several repetitions of the enclosed components. Curly brackets with a trailing plus sign {
}+ denote one or several repetitions of the enclosed components. Parentheses (
) denote grouping.

11.1 Lexical conventions

Blanks

The following characters are considered as blanks: space, horizontal tabulation, carriage return, line feed and form feed. Blanks are ignored, but they separate adjacent identifiers, literals and keywords that would otherwise be confused as one single identifier, literal or keyword.

Comments

Comments are introduced by the two characters (*, with no intervening blanks, and terminated by the characters *), with no intervening blanks. Comments are treated as blank characters. Comments do not occur inside string or character literals. Nested comments are handled correctly.

(* single line comment *) (* multiple line comment, commenting out part of a program, and containing a nested comment: let f = function | 'A'..'Z' -> "Uppercase" (* Add other cases later... *) *)

Identifiers

ident::= (letter ∣ _) { letter ∣ 0
9 ∣ _ ∣ ' } 
 
capitalized-ident::= (A
Z) { letter ∣ 0
9 ∣ _ ∣ ' } 
 
lowercase-ident::= (a
z ∣ _) { letter ∣ 0
9 ∣ _ ∣ ' } 
 
letter::= A
Z ∣ a
z

Identifiers are sequences of letters, digits, _ (the underscore character), and ' (the single quote), starting with a letter or an underscore. Letters contain at least the 52 lowercase and uppercase letters from the ASCII set. The current implementation also recognizes as letters some characters from the ISO 8859-1 set (characters 192–214 and 216–222 as uppercase letters; characters 223–246 and 248–255 as lowercase letters). This feature is deprecated and should be avoided for future compatibility.

All characters in an identifier are meaningful. The current implementation accepts identifiers up to 16000000 characters in length.

In many places, OCaml makes a distinction between capitalized identifiers and identifiers that begin with a lowercase letter. The underscore character is considered a lowercase letter for this purpose.

Integer literals

integer-literal::= [-] (0
9) { 0
9 ∣ _ }
 âˆŁâ€„[-] (0x ∣ 0X) (0
9 ∣ A
F ∣ a
f) { 0
9 ∣ A
F ∣ a
f ∣ _ }
 âˆŁâ€„[-] (0o ∣ 0O) (0
7) { 0
7 ∣ _ }
 âˆŁâ€„[-] (0b ∣ 0B) (0
1) { 0
1 ∣ _ }
 
int32-literal::= integer-literal l
 
int64-literal::= integer-literal L
 
nativeint-literal::= integer-literal n

An integer literal is a sequence of one or more digits, optionally preceded by a minus sign. By default, integer literals are in decimal (radix 10). The following prefixes select a different radix:

PrefixRadix
0x, 0Xhexadecimal (radix 16)
0o, 0Ooctal (radix 8)
0b, 0Bbinary (radix 2)

(The initial 0 is the digit zero; the O for octal is the letter O.) An integer literal can be followed by one of the letters l, L or n to indicate that this integer has type int32, int64 or nativeint respectively, instead of the default type int for integer literals. The interpretation of integer literals that fall outside the range of representable integer values is undefined.

For convenience and readability, underscore characters (_) are accepted (and ignored) within integer literals.

# let house_number = 37 let million = 1_000_000 let copyright = 0x00A9 let counter64bit = ref 0L;;
val house_number : int = 37 val million : int = 1000000 val copyright : int = 169 val counter64bit : int64 ref = {contents = 0L}

Floating-point literals

float-literal::= [-] (0
9) { 0
9 ∣ _ } [. { 0
9 ∣ _ }] [(e ∣ E) [+ ∣ -] (0
9) { 0
9 ∣ _ }]
 âˆŁâ€„[-] (0x ∣ 0X) (0
9 ∣ A
F ∣ a
f) { 0
9 ∣ A
F ∣ a
f ∣ _ }  [. { 0
9 ∣ A
F ∣ a
f ∣ _ }] [(p ∣ P) [+ ∣ -] (0
9) { 0
9 ∣ _ }]

Floating-point decimal literals consist in an integer part, a fractional part and an exponent part. The integer part is a sequence of one or more digits, optionally preceded by a minus sign. The fractional part is a decimal point followed by zero, one or more digits. The exponent part is the character e or E followed by an optional + or - sign, followed by one or more digits. It is interpreted as a power of 10. The fractional part or the exponent part can be omitted but not both, to avoid ambiguity with integer literals. The interpretation of floating-point literals that fall outside the range of representable floating-point values is undefined.

Floating-point hexadecimal literals are denoted with the 0x or 0X prefix. The syntax is similar to that of floating-point decimal literals, with the following differences. The integer part and the fractional part use hexadecimal digits. The exponent part starts with the character p or P. It is written in decimal and interpreted as a power of 2.

For convenience and readability, underscore characters (_) are accepted (and ignored) within floating-point literals.

# let pi = 3.141_592_653_589_793_12 let small_negative = -1e-5 let machine_epsilon = 0x1p-52;;
val pi : float = 3.14159265358979312 val small_negative : float = -1e-05 val machine_epsilon : float = 2.22044604925031308e-16

Character literals

char-literal::= ' regular-char '
 âˆŁâ€„' escape-sequence '
 
escape-sequence::= \ (\ ∣ " ∣ ' ∣ n ∣ t ∣ b ∣ r ∣ space)
 âˆŁâ€„\ (0
9) (0
9) (0
9)
 âˆŁâ€„\x (0
9 ∣ A
F ∣ a
f) (0
9 ∣ A
F ∣ a
f)
 âˆŁâ€„\o (0
3) (0
7) (0
7)

Character literals are delimited by ' (single quote) characters. The two single quotes enclose either one character different from ' and \, or one of the escape sequences below:

SequenceCharacter denoted
\\backslash (\)
\"double quote (")
\'single quote (')
\nlinefeed (LF)
\rcarriage return (CR)
\thorizontal tabulation (TAB)
\bbackspace (BS)
\spacespace (SPC)
\dddthe character with ASCII code ddd in decimal
\xhhthe character with ASCII code hh in hexadecimal
\oooothe character with ASCII code ooo in octal
# let a = 'a' let single_quote = '\'' let copyright = '\xA9';;
val a : char = 'a' val single_quote : char = '\'' val copyright : char = '\169'

String literals

string-literal::= " { string-character } "
 âˆŁâ€„ { quoted-string-id | { any-char } | quoted-string-id }
 
quoted-string-id::= { a...z ∣ _ }
 
string-character::= regular-string-char
 âˆŁâ€„escape-sequence
 âˆŁâ€„\u{ { 0
9 ∣ A
F ∣ a
f }+ }
 âˆŁâ€„\ newline { space ∣ tab }

String literals are delimited by " (double quote) characters. The two double quotes enclose a sequence of either characters different from " and \, or escape sequences from the table given above for character literals, or a Unicode character escape sequence.

A Unicode character escape sequence is substituted by the UTF-8 encoding of the specified Unicode scalar value. The Unicode scalar value, an integer in the ranges 0x0000...0xD7FF or 0xE000...0x10FFFF, is defined using 1 to 6 hexadecimal digits; leading zeros are allowed.

# let greeting = "Hello, World!\n" let superscript_plus = "\u{207A}";;
val greeting : string = "Hello, World!\n" val superscript_plus : string = "âș"

To allow splitting long string literals across lines, the sequence \newline spaces-or-tabs (a backslash at the end of a line followed by any number of spaces and horizontal tabulations at the beginning of the next line) is ignored inside string literals.

# let longstr = "Call me Ishmael. Some years ago — never mind how long \ precisely — having little or no money in my purse, and \ nothing particular to interest me on shore, I thought I\ \ would sail about a little and see the watery part of t\ he world.";;
val longstr : string = "Call me Ishmael. Some years ago — never mind how long precisely — having little or no money in my purse, and nothing particular to interest me on shore, I thought I would sail about a little and see the watery part of the world."

Quoted string literals provide an alternative lexical syntax for string literals. They are useful to represent strings of arbitrary content without escaping. Quoted strings are delimited by a matching pair of { quoted-string-id | and | quoted-string-id } with the same quoted-string-id on both sides. Quoted strings do not interpret any character in a special way but requires that the sequence | quoted-string-id } does not occur in the string itself. The identifier quoted-string-id is a (possibly empty) sequence of lowercase letters and underscores that can be freely chosen to avoid such issue.

# let quoted_greeting = {|"Hello, World!"|} let nested = {ext|hello {|world|}|ext};;
val quoted_greeting : string = "\"Hello, World!\"" val nested : string = "hello {|world|}"

The current implementation places practically no restrictions on the length of string literals.

Naming labels

To avoid ambiguities, naming labels in expressions cannot just be defined syntactically as the sequence of the three tokens ~, ident and :, and have to be defined at the lexical level.

label-name::= lowercase-ident
 
label::= ~ label-name :
 
optlabel::= ? label-name :

Naming labels come in two flavours: label for normal arguments and optlabel for optional ones. They are simply distinguished by their first character, either ~ or ?.

Despite label and optlabel being lexical entities in expressions, their expansions ~ label-name : and ? label-name : will be used in grammars, for the sake of readability. Note also that inside type expressions, this expansion can be taken literally, i.e. there are really 3 tokens, with optional blanks between them.

Prefix and infix symbols

infix-symbol::= (core-operator-char ∣ % ∣ <) { operator-char }
 âˆŁâ€„# { operator-char }+
 
prefix-symbol::= ! { operator-char }
 âˆŁâ€„(? ∣ ~) { operator-char }+
 
operator-char::= ~ ∣ ! ∣ ? ∣ core-operator-char ∣ % ∣ < ∣ : ∣ .
 
core-operator-char::= $ ∣ & ∣ * ∣ + ∣ - ∣ / ∣ = ∣ > ∣ @ ∣ ^ ∣ |

See also the following language extensions: extension operators, extended indexing operators, and binding operators.

Sequences of “operator characters”, such as <=> or !!, are read as a single token from the infix-symbol or prefix-symbol class. These symbols are parsed as prefix and infix operators inside expressions, but otherwise behave like normal identifiers.

Keywords

The identifiers below are reserved as keywords, and cannot be employed otherwise:

      and         as          assert      asr         begin       class
      constraint  do          done        downto      else        end
      exception   external    false       for         fun         function
      functor     if          in          include     inherit     initializer
      land        lazy        let         lor         lsl         lsr
      lxor        match       method      mod         module      mutable
      new         nonrec      object      of          open        or
      private     rec         sig         struct      then        to
      true        try         type        val         virtual     when
      while       with


The following character sequences are also keywords:

    !=    #     &     &&    '     (     )     *     +     ,     -
    -.    ->    .     ..    .~    :     ::    :=    :>    ;     ;;
    <     <-    =     >     >]    >}    ?     [     [<    [>    [|
    ]     _     `     {     {<    |     |]    ||    }     ~

Note that the following identifiers are keywords of the now unmaintained Camlp4 system and should be avoided for backwards compatibility reasons.

    parser    value    $     $$    $:    <:    <<    >>    ??

Ambiguities

Lexical ambiguities are resolved according to the “longest match” rule: when a character sequence can be decomposed into two tokens in several different ways, the decomposition retained is the one with the longest first token.

Line number directives

linenum-directive::= # { 0
9 }+ " { string-character } "

Preprocessors that generate OCaml source code can insert line number directives in their output so that error messages produced by the compiler contain line numbers and file names referring to the source file before preprocessing, instead of after preprocessing. A line number directive starts at the beginning of a line, is composed of a # (sharp sign), followed by a positive integer (the source line number), followed by a character string (the source file name). Line number directives are treated as blanks during lexical analysis.

11.2 Values

This section describes the kinds of values that are manipulated by OCaml programs.

11.2.1 Base values

Integer numbers

Integer values are integer numbers from −230 to 230−1, that is −1073741824 to 1073741823. The implementation may support a wider range of integer values: on 64-bit platforms, the current implementation supports integers ranging from −262 to 262−1.

Floating-point numbers

Floating-point values are numbers in floating-point representation. The current implementation uses double-precision floating-point numbers conforming to the IEEE 754 standard, with 53 bits of mantissa and an exponent ranging from −1022 to 1023.

Characters

Character values are represented as 8-bit integers between 0 and 255. Character codes between 0 and 127 are interpreted following the ASCII standard. The current implementation interprets character codes between 128 and 255 following the ISO 8859-1 standard.

Character strings

String values are finite sequences of characters. The current implementation supports strings containing up to 224 − 5 characters (16777211 characters); on 64-bit platforms, the limit is 257 − 9.

11.2.2 Tuples

Tuples of values are written (v1, 
, vn), standing for the n-tuple of values v1 to vn. The current implementation supports tuple of up to 222 − 1 elements (4194303 elements).

11.2.3 Records

Record values are labeled tuples of values. The record value written { field1 = v1; 
; fieldn = vn } associates the value vi to the record field fieldi, for i = 1 
 n. The current implementation supports records with up to 222 − 1 fields (4194303 fields).

11.2.4 Arrays

Arrays are finite, variable-sized sequences of values of the same type. The current implementation supports arrays containing up to 222 − 1 elements (4194303 elements) unless the elements are floating-point numbers (2097151 elements in this case); on 64-bit platforms, the limit is 254 − 1 for all arrays.

11.2.5 Variant values

Variant values are either a constant constructor, or a non-constant constructor applied to a number of values. The former case is written constr; the latter case is written constr (v1, ... , vn ), where the vi are said to be the arguments of the non-constant constructor constr. The parentheses may be omitted if there is only one argument.

The following constants are treated like built-in constant constructors:

ConstantConstructor
falsethe boolean false
truethe boolean true
()the “unit” value
[]the empty list

The current implementation limits each variant type to have at most 246 non-constant constructors and 230−1 constant constructors.

11.2.6 Polymorphic variants

Polymorphic variants are an alternate form of variant values, not belonging explicitly to a predefined variant type, and following specific typing rules. They can be either constant, written `tag-name, or non-constant, written `tag-name(v).

11.2.7 Functions

Functional values are mappings from values to values.

11.2.8 Objects

Objects are composed of a hidden internal state which is a record of instance variables, and a set of methods for accessing and modifying these variables. The structure of an object is described by the toplevel class that created it.

11.3 Names

Identifiers are used to give names to several classes of language objects and refer to these objects by name later:

These eleven name spaces are distinguished both by the context and by the capitalization of the identifier: whether the first letter of the identifier is in lowercase (written lowercase-ident below) or in uppercase (written capitalized-ident). Underscore is considered a lowercase letter for this purpose.

Naming objects

value-name::= lowercase-ident
 âˆŁâ€„( operator-name )
 
operator-name::= prefix-symbol ∣ infix-op
 
infix-op::= infix-symbol
 âˆŁâ€„* ∣ + ∣ - ∣ -. ∣ = ∣ != ∣ < ∣ > ∣ or ∣ || ∣ & ∣ && ∣ :=
 âˆŁâ€„mod ∣ land ∣ lor ∣ lxor ∣ lsl ∣ lsr ∣ asr
 
constr-name::= capitalized-ident
 
tag-name::= capitalized-ident
 
typeconstr-name::= lowercase-ident
 
field-name::= lowercase-ident
 
module-name::= capitalized-ident
 
modtype-name::= ident
 
class-name::= lowercase-ident
 
inst-var-name::= lowercase-ident
 
method-name::= lowercase-ident

See also the following language extension: extended indexing operators.

As shown above, prefix and infix symbols as well as some keywords can be used as value names, provided they are written between parentheses. The capitalization rules are summarized in the table below.

Name spaceCase of first letter
Valueslowercase
Constructorsuppercase
Labelslowercase
Polymorphic variant tagsuppercase
Exceptionsuppercase
Type constructorslowercase
Record fieldslowercase
Classeslowercase
Instance variableslowercase
Methodslowercase
Modulesuppercase
Module typesany

Note on polymorphic variant tags: the current implementation accepts lowercase variant tags in addition to capitalized variant tags, but we suggest you avoid lowercase variant tags for portability and compatibility with future OCaml versions.

Referring to named objects

value-path::= [ module-path . ] value-name
 
constr::= [ module-path . ] constr-name
 
typeconstr::= [ extended-module-path . ] typeconstr-name
 
field::= [ module-path . ] field-name
 
modtype-path::= [ extended-module-path . ] modtype-name
 
class-path::= [ module-path . ] class-name
 
classtype-path::= [ extended-module-path . ] class-name
 
module-path::= module-name { . module-name }
 
extended-module-path::= extended-module-name { . extended-module-name }
 
extended-module-name::= module-name { ( extended-module-path ) }

A named object can be referred to either by its name (following the usual static scoping rules for names) or by an access path prefix . name, where prefix designates a module and name is the name of an object defined in that module. The first component of the path, prefix, is either a simple module name or an access path name1 . name2 
, in case the defining module is itself nested inside other modules. For referring to type constructors, module types, or class types, the prefix can also contain simple functor applications (as in the syntactic class extended-module-path above) in case the defining module is the result of a functor application.

Label names, tag names, method names and instance variable names need not be qualified: the former three are global labels, while the latter are local to a class.

11.4 Type expressions

typexpr::= ' ident
 âˆŁâ€„_
 âˆŁâ€„( typexpr )
 âˆŁâ€„[[?]label-name:] typexpr -> typexpr
 âˆŁâ€„typexpr { * typexpr }+
 âˆŁâ€„typeconstr
 âˆŁâ€„typexpr typeconstr
 âˆŁâ€„( typexpr { , typexpr } ) typeconstr
 âˆŁâ€„typexpr as ' ident
 âˆŁâ€„polymorphic-variant-type
 âˆŁâ€„< [..] >
 âˆŁâ€„< method-type { ; method-type } [; ∣ ; ..] >
 âˆŁâ€„# classtype-path
 âˆŁâ€„typexpr # class-path
 âˆŁâ€„( typexpr { , typexpr } ) # class-path
 
poly-typexpr::= typexpr
 âˆŁâ€„{ ' ident }+ . typexpr
 
method-type::= method-name : poly-typexpr

See also the following language extensions: first-class modules, attributes and extension nodes.

The table below shows the relative precedences and associativity of operators and non-closed type constructions. The constructions with higher precedences come first.

OperatorAssociativity
Type constructor application–
#–
*–
->right
as–

Type expressions denote types in definitions of data types as well as in type constraints over patterns and expressions.

Type variables

The type expression ' ident stands for the type variable named ident. The type expression _ stands for either an anonymous type variable or anonymous type parameters. In data type definitions, type variables are names for the data type parameters. In type constraints, they represent unspecified types that can be instantiated by any type to satisfy the type constraint. In general the scope of a named type variable is the whole top-level phrase where it appears, and it can only be generalized when leaving this scope. Anonymous variables have no such restriction. In the following cases, the scope of named type variables is restricted to the type expression where they appear: 1) for universal (explicitly polymorphic) type variables; 2) for type variables that only appear in public method specifications (as those variables will be made universal, as described in section 11.9.1); 3) for variables used as aliases, when the type they are aliased to would be invalid in the scope of the enclosing definition (i.e. when it contains free universal type variables, or locally defined types.)

Parenthesized types

The type expression ( typexpr ) denotes the same type as typexpr.

Function types

The type expression typexpr1 -> typexpr2 denotes the type of functions mapping arguments of type typexpr1 to results of type typexpr2.

label-name : typexpr1 -> typexpr2 denotes the same function type, but the argument is labeled label.

? label-name : typexpr1 -> typexpr2 denotes the type of functions mapping an optional labeled argument of type typexpr1 to results of type typexpr2. That is, the physical type of the function will be typexpr1 option -> typexpr2.

Tuple types

The type expression typexpr1 * 
 * typexprn denotes the type of tuples whose elements belong to types typexpr1, 
 typexprn respectively.

Constructed types

Type constructors with no parameter, as in typeconstr, are type expressions.

The type expression typexpr typeconstr, where typeconstr is a type constructor with one parameter, denotes the application of the unary type constructor typeconstr to the type typexpr.

The type expression (typexpr1,
,typexprn) typeconstr, where typeconstr is a type constructor with n parameters, denotes the application of the n-ary type constructor typeconstr to the types typexpr1 through typexprn.

In the type expression _ typeconstr , the anonymous type expression _ stands in for anonymous type parameters and is equivalent to (_, 
,_) with as many repetitions of _ as the arity of typeconstr.

Aliased and recursive types

The type expression typexpr as ' ident denotes the same type as typexpr, and also binds the type variable ident to type typexpr both in typexpr and in other types. In general the scope of an alias is the same as for a named type variable, and covers the whole enclosing definition. If the type variable ident actually occurs in typexpr, a recursive type is created. Recursive types for which there exists a recursive path that does not contain an object or polymorphic variant type constructor are rejected, except when the -rectypes mode is selected.

If ' ident denotes an explicit polymorphic variable, and typexpr denotes either an object or polymorphic variant type, the row variable of typexpr is captured by ' ident, and quantified upon.

Polymorphic variant types

polymorphic-variant-type::= [ tag-spec-first { | tag-spec } ]
 âˆŁâ€„[> [ tag-spec ] { | tag-spec } ]
 âˆŁâ€„[< [|] tag-spec-full { | tag-spec-full } [ > { `tag-name }+ ] ]
 
tag-spec-first::= `tag-name [ of typexpr ]
 âˆŁâ€„[ typexpr ] | tag-spec
 
tag-spec::= `tag-name [ of typexpr ]
 âˆŁâ€„typexpr
 
tag-spec-full::= `tag-name [ of [&] typexpr { & typexpr } ]
 âˆŁâ€„typexpr

Polymorphic variant types describe the values a polymorphic variant may take.

The first case is an exact variant type: all possible tags are known, with their associated types, and they can all be present. Its structure is fully known.

The second case is an open variant type, describing a polymorphic variant value: it gives the list of all tags the value could take, with their associated types. This type is still compatible with a variant type containing more tags. A special case is the unknown type, which does not define any tag, and is compatible with any variant type.

The third case is a closed variant type. It gives information about all the possible tags and their associated types, and which tags are known to potentially appear in values. The exact variant type (first case) is just an abbreviation for a closed variant type where all possible tags are also potentially present.

In all three cases, tags may be either specified directly in the `tag-name [of typexpr] form, or indirectly through a type expression, which must expand to an exact variant type, whose tag specifications are inserted in its place.

Full specifications of variant tags are only used for non-exact closed types. They can be understood as a conjunctive type for the argument: it is intended to have all the types enumerated in the specification.

Such conjunctive constraints may be unsatisfiable. In such a case the corresponding tag may not be used in a value of this type. This does not mean that the whole type is not valid: one can still use other available tags. Conjunctive constraints are mainly intended as output from the type checker. When they are used in source programs, unsolvable constraints may cause early failures.

Object types

An object type < [method-type { ; method-type }] > is a record of method types.

Each method may have an explicit polymorphic type: { ' ident }+ . typexpr. Explicit polymorphic variables have a local scope, and an explicit polymorphic type can only be unified to an equivalent one, where only the order and names of polymorphic variables may change.

The type < { method-type ; } .. > is the type of an object whose method names and types are described by method-type1, 
, method-typen, and possibly some other methods represented by the ellipsis. This ellipsis actually is a special kind of type variable (called row variable in the literature) that stands for any number of extra method types.

#-types

The type # classtype-path is a special kind of abbreviation. This abbreviation unifies with the type of any object belonging to a subclass of the class type classtype-path. It is handled in a special way as it usually hides a type variable (an ellipsis, representing the methods that may be added in a subclass). In particular, it vanishes when the ellipsis gets instantiated. Each type expression # classtype-path defines a new type variable, so type # classtype-path -> # classtype-path is usually not the same as type (# classtype-path as ' ident) -> ' ident.

Variant and record types

There are no type expressions describing (defined) variant types nor record types, since those are always named, i.e. defined before use and referred to by name. Type definitions are described in section 11.8.1.

11.5 Constants

constant::= integer-literal
 âˆŁâ€„int32-literal
 âˆŁâ€„int64-literal
 âˆŁâ€„nativeint-literal
 âˆŁâ€„float-literal
 âˆŁâ€„char-literal
 âˆŁâ€„string-literal
 âˆŁâ€„constr
 âˆŁâ€„false
 âˆŁâ€„true
 âˆŁâ€„()
 âˆŁâ€„begin end
 âˆŁâ€„[]
 âˆŁâ€„[||]
 âˆŁâ€„`tag-name

See also the following language extension: extension literals.

The syntactic class of constants comprises literals from the four base types (integers, floating-point numbers, characters, character strings), the integer variants, and constant constructors from both normal and polymorphic variants, as well as the special constants false, true, (), [], and [||], which behave like constant constructors, and begin end, which is equivalent to ().

11.6 Patterns

pattern::= value-name
 âˆŁâ€„_
 âˆŁâ€„constant
 âˆŁâ€„pattern as value-name
 âˆŁâ€„( pattern )
 âˆŁâ€„( pattern : typexpr )
 âˆŁâ€„pattern | pattern
 âˆŁâ€„constr pattern
 âˆŁâ€„`tag-name pattern
 âˆŁâ€„#typeconstr
 âˆŁâ€„pattern { , pattern }+
 âˆŁâ€„{ field [: typexpr] [= pattern]{ ; field [: typexpr] [= pattern] } [; _ ] [ ; ] }
 âˆŁâ€„[ pattern { ; pattern } [ ; ] ]
 âˆŁâ€„pattern :: pattern
 âˆŁâ€„[| pattern { ; pattern } [ ; ] |]
 âˆŁâ€„char-literal .. char-literal
 âˆŁâ€„lazy pattern
 âˆŁâ€„exception pattern
 âˆŁâ€„module-path .( pattern )
 âˆŁâ€„module-path .[ pattern ]
 âˆŁâ€„module-path .[| pattern |]
 âˆŁâ€„module-path .{ pattern }

See also the following language extensions: first-class modules, attributes and extension nodes.

The table below shows the relative precedences and associativity of operators and non-closed pattern constructions. The constructions with higher precedences come first.

OperatorAssociativity
..–
lazy (see section 11.6)–
Constructor application, Tag applicationright
::right
,–
|left
as–

Patterns are templates that allow selecting data structures of a given shape, and binding identifiers to components of the data structure. This selection operation is called pattern matching; its outcome is either “this value does not match this pattern”, or “this value matches this pattern, resulting in the following bindings of names to values”.

Variable patterns

A pattern that consists in a value name matches any value, binding the name to the value. The pattern _ also matches any value, but does not bind any name.

# let is_empty = function | [] -> true | _ :: _ -> false;;
val is_empty : 'a list -> bool = <fun>

Patterns are linear: a variable cannot be bound several times by a given pattern. In particular, there is no way to test for equality between two parts of a data structure using only a pattern:

# let pair_equal = function | x, x -> true | x, y -> false;;
Error: Variable x is bound several times in this matching

However, we can use a when guard for this purpose:

# let pair_equal = function | x, y when x = y -> true | _ -> false;;
val pair_equal : 'a * 'a -> bool = <fun>

Constant patterns

A pattern consisting in a constant matches the values that are equal to this constant.

# let bool_of_string = function | "true" -> true | "false" -> false | _ -> raise (Invalid_argument "bool_of_string");;
val bool_of_string : string -> bool = <fun>

Alias patterns

The pattern pattern1 as value-name matches the same values as pattern1. If the matching against pattern1 is successful, the name value-name is bound to the matched value, in addition to the bindings performed by the matching against pattern1.

# let sort_pair ((x, y) as p) = if x <= y then p else (y, x);;
val sort_pair : 'a * 'a -> 'a * 'a = <fun>

Parenthesized patterns

The pattern ( pattern1 ) matches the same values as pattern1. A type constraint can appear in a parenthesized pattern, as in ( pattern1 : typexpr ). This constraint forces the type of pattern1 to be compatible with typexpr.

# let int_triple_is_ordered ((a, b, c) : int * int * int) = a <= b && b <= c;;
val int_triple_is_ordered : int * int * int -> bool = <fun>

“Or” patterns

The pattern pattern1 | pattern2 represents the logical “or” of the two patterns pattern1 and pattern2. A value matches pattern1 | pattern2 if it matches pattern1 or pattern2. The two sub-patterns pattern1 and pattern2 must bind exactly the same identifiers to values having the same types. Matching is performed from left to right. More precisely, in case some value v matches pattern1 | pattern2, the bindings performed are those of pattern1 when v matches pattern1. Otherwise, value v matches pattern2 whose bindings are performed.

# type shape = Square of float | Rect of (float * float) | Circle of float let is_rectangular = function | Square _ | Rect _ -> true | Circle _ -> false;;
type shape = Square of float | Rect of (float * float) | Circle of float val is_rectangular : shape -> bool = <fun>

Variant patterns

The pattern constr ( pattern1 , 
 , patternn ) matches all variants whose constructor is equal to constr, and whose arguments match pattern1 
 patternn. It is a type error if n is not the number of arguments expected by the constructor.

The pattern constr _ matches all variants whose constructor is constr.

# type 'a tree = Lf | Br of 'a tree * 'a * 'a tree let rec total = function | Br (l, x, r) -> total l + x + total r | Lf -> 0;;
type 'a tree = Lf | Br of 'a tree * 'a * 'a tree val total : int tree -> int = <fun>

The pattern pattern1 :: pattern2 matches non-empty lists whose heads match pattern1, and whose tails match pattern2.

The pattern [ pattern1 ; 
 ; patternn ] matches lists of length n whose elements match pattern1 
patternn, respectively. This pattern behaves like pattern1 :: 
 :: patternn :: [].

# let rec destutter = function | [] -> [] | [a] -> [a] | a :: b :: t -> if a = b then destutter (b :: t) else a :: destutter (b :: t);;
val destutter : 'a list -> 'a list = <fun>

Polymorphic variant patterns

The pattern `tag-name pattern1 matches all polymorphic variants whose tag is equal to tag-name, and whose argument matches pattern1.

# let rec split = function | [] -> ([], []) | h :: t -> let ss, gs = split t in match h with | `Sheep _ as s -> (s :: ss, gs) | `Goat _ as g -> (ss, g :: gs);;
val split : [< `Goat of 'a | `Sheep of 'b ] list -> [> `Sheep of 'b ] list * [> `Goat of 'a ] list = <fun>

Polymorphic variant abbreviation patterns

If the type [('a,'b,
)] typeconstr = [ `tag-name1 typexpr1 | 
 | `tag-namen typexprn] is defined, then the pattern #typeconstr is a shorthand for the following or-pattern: ( `tag-name1(_ : typexpr1) | 
 | `tag-namen(_ : typexprn)). It matches all values of type [< typeconstr ].

# type 'a rectangle = [`Square of 'a | `Rectangle of 'a * 'a] type 'a shape = [`Circle of 'a | 'a rectangle] let try_rectangle = function | #rectangle as r -> Some r | `Circle _ -> None;;
type 'a rectangle = [ `Rectangle of 'a * 'a | `Square of 'a ] type 'a shape = [ `Circle of 'a | `Rectangle of 'a * 'a | `Square of 'a ] val try_rectangle : [< `Circle of 'a | `Rectangle of 'b * 'b | `Square of 'b ] -> [> `Rectangle of 'b * 'b | `Square of 'b ] option = <fun>

Tuple patterns

The pattern pattern1 , 
 , patternn matches n-tuples whose components match the patterns pattern1 through patternn. That is, the pattern matches the tuple values (v1, 
, vn) such that patterni matches vi for i = 1,
 , n.

# let vector (x0, y0) (x1, y1) = (x1 -. x0, y1 -. y0);;
val vector : float * float -> float * float -> float * float = <fun>

Record patterns

The pattern { field1 [= pattern1] ; 
 ; fieldn [= patternn] } matches records that define at least the fields field1 through fieldn, and such that the value associated to fieldi matches the pattern patterni, for i = 1,
 , n. A single identifier fieldk stands for fieldk = fieldk , and a single qualified identifier module-path . fieldk stands for module-path . fieldk = fieldk . The record value can define more fields than field1 
fieldn; the values associated to these extra fields are not taken into account for matching. Optionally, a record pattern can be terminated by ; _ to convey the fact that not all fields of the record type are listed in the record pattern and that it is intentional. Optional type constraints can be added field by field with { field1 : typexpr1 = pattern1 ;
 ;fieldn : typexprn = patternn } to force the type of fieldk to be compatible with typexprk.

# let bytes_allocated {Gc.minor_words = minor; Gc.major_words = major; Gc.promoted_words = prom; _} = (Sys.word_size / 4) * int_of_float (minor +. major -. prom);;
val bytes_allocated : Gc.stat -> int = <fun>

Array patterns

The pattern [| pattern1 ; 
 ; patternn |] matches arrays of length n such that the i-th array element matches the pattern patterni, for i = 1,
 , n.

# let matrix3_is_symmetric = function | [|[|_; b; c|]; [|d; _; f|]; [|g; h; _|]|] -> b = d && c = g && f = h | _ -> failwith "matrix3_is_symmetric: not a 3x3 matrix";;
val matrix3_is_symmetric : 'a array array -> bool = <fun>

Range patterns

The pattern ' c ' .. ' d ' is a shorthand for the pattern

' c ' | ' c1 ' | ' c2 ' | 
 | ' cn ' | ' d '

where c1, c2, 
, cn are the characters that occur between c and d in the ASCII character set. For instance, the pattern '0'..'9' matches all characters that are digits.

# type char_class = Uppercase | Lowercase | Digit | Other let classify_char = function | 'A'..'Z' -> Uppercase | 'a'..'z' -> Lowercase | '0'..'9' -> Digit | _ -> Other;;
type char_class = Uppercase | Lowercase | Digit | Other val classify_char : char -> char_class = <fun>

Lazy patterns

(Introduced in Objective Caml 3.11)

pattern::= ...

The pattern lazy pattern matches a value v of type Lazy.t, provided pattern matches the result of forcing v with Lazy.force. A successful match of a pattern containing lazy sub-patterns forces the corresponding parts of the value being matched, even those that imply no test such as lazy value-name or lazy _. Matching a value with a pattern-matching where some patterns contain lazy sub-patterns may imply forcing parts of the value, even when the pattern selected in the end has no lazy sub-pattern.

# let force_opt = function | Some (lazy n) -> n | None -> 0;;
val force_opt : int lazy_t option -> int = <fun>

For more information, see the description of module Lazy in the standard library (module Lazy).

Exception patterns

(Introduced in OCaml 4.02)

A new form of exception pattern, exception pattern , is allowed only as a toplevel pattern or inside a toplevel or-pattern under a match...with pattern-matching (other occurrences are rejected by the type-checker).

Cases with such a toplevel pattern are called “exception cases”, as opposed to regular “value cases”. Exception cases are applied when the evaluation of the matched expression raises an exception. The exception value is then matched against all the exception cases and re-raised if none of them accept the exception (as with a try...with block). Since the bodies of all exception and value cases are outside the scope of the exception handler, they are all considered to be in tail-position: if the match...with block itself is in tail position in the current function, any function call in tail position in one of the case bodies results in an actual tail call.

A pattern match must contain at least one value case. It is an error if all cases are exceptions, because there would be no code to handle the return of a value.

# let find_opt p l = match List.find p l with | exception Not_found -> None | x -> Some x;;
val find_opt : ('a -> bool) -> 'a list -> 'a option = <fun>

Local opens for patterns

(Introduced in OCaml 4.04)

For patterns, local opens are limited to the module-path.(pattern) construction. This construction locally opens the module referred to by the module path module-path in the scope of the pattern pattern.

When the body of a local open pattern is delimited by [ ], [| |], or { }, the parentheses can be omitted. For example, module-path.[pattern] is equivalent to module-path.([pattern]), and module-path.[| pattern |] is equivalent to module-path.([| pattern |]).

# let bytes_allocated Gc.{minor_words; major_words; promoted_words; _} = (Sys.word_size / 4) * int_of_float (minor_words +. major_words -. promoted_words);;
val bytes_allocated : Gc.stat -> int = <fun>

11.7 Expressions

expr::= value-path
 âˆŁâ€„constant
 âˆŁâ€„( expr )
 âˆŁâ€„begin expr end
 âˆŁâ€„( expr : typexpr )
 âˆŁâ€„expr { , expr }+
 âˆŁâ€„constr expr
 âˆŁâ€„`tag-name expr
 âˆŁâ€„expr :: expr
 âˆŁâ€„[ expr { ; expr } [;] ]
 âˆŁâ€„[| expr { ; expr } [;] |]
 âˆŁâ€„{ field [: typexpr] [= expr]{ ; field [: typexpr] [= expr] } [;] }
 âˆŁâ€„{ expr with field [: typexpr] [= expr]{ ; field [: typexpr] [= expr] } [;] }
 âˆŁâ€„expr { argument }+
 âˆŁâ€„prefix-symbol expr
 âˆŁâ€„- expr
 âˆŁâ€„-. expr
 âˆŁâ€„expr infix-op expr
 âˆŁâ€„expr . field
 âˆŁâ€„expr . field <- expr
 âˆŁâ€„expr .( expr )
 âˆŁâ€„expr .( expr ) <- expr
 âˆŁâ€„expr .[ expr ]
 âˆŁâ€„expr .[ expr ] <- expr
 âˆŁâ€„if expr then expr [ else expr ]
 âˆŁâ€„while expr do expr done
 âˆŁâ€„for value-name = expr ( to ∣ downto ) expr do expr done
 âˆŁâ€„expr ; expr
 âˆŁâ€„match expr with pattern-matching
 âˆŁâ€„function pattern-matching
 âˆŁâ€„fun { parameter }+ [ : typexpr ] -> expr
 âˆŁâ€„try expr with pattern-matching
 âˆŁâ€„let [rec] let-binding { and let-binding } in expr
 âˆŁâ€„let exception constr-decl in expr
 âˆŁâ€„let module module-name { ( module-name : module-type ) } [ : module-type ]  = module-expr in expr
 âˆŁâ€„( expr :> typexpr )
 âˆŁâ€„( expr : typexpr :> typexpr )
 âˆŁâ€„assert expr
 âˆŁâ€„lazy expr
 âˆŁâ€„local-open
 âˆŁâ€„object-expr
 
argument::= expr
 âˆŁâ€„~ label-name
 âˆŁâ€„~ label-name : expr
 âˆŁâ€„? label-name
 âˆŁâ€„? label-name : expr
 
pattern-matching::= [ | ] pattern [when expr] -> expr { | pattern [when expr] -> expr }
 
let-binding::= pattern = expr
 âˆŁâ€„value-name { parameter } [: typexpr] [:> typexpr] = expr
 âˆŁâ€„value-name : poly-typexpr = expr 
 
parameter::= pattern
 âˆŁâ€„~ label-name
 âˆŁâ€„~ ( label-name [: typexpr] )
 âˆŁâ€„~ label-name : pattern
 âˆŁâ€„? label-name
 âˆŁâ€„? ( label-name [: typexpr] [= expr] )
 âˆŁâ€„? label-name : pattern
 âˆŁâ€„? label-name : ( pattern [: typexpr] [= expr] )
 
local-open::= 
 âˆŁâ€„let open module-path in expr
 âˆŁâ€„module-path .( expr )
 âˆŁâ€„module-path .[ expr ]
 âˆŁâ€„module-path .[| expr |]
 âˆŁâ€„module-path .{ expr }
 âˆŁâ€„module-path .{< expr >}
 
object-expr::= 
 âˆŁâ€„new class-path
 âˆŁâ€„object class-body end
 âˆŁâ€„expr # method-name
 âˆŁâ€„inst-var-name
 âˆŁâ€„inst-var-name <- expr
 âˆŁâ€„{< [ inst-var-name [= expr] { ; inst-var-name [= expr] } [;] ] >}

See also the following language extensions: first-class modules, overriding in open statements, syntax for Bigarray access, attributes, extension nodes and extended indexing operators.

11.7.1 Precedence and associativity

The table below shows the relative precedences and associativity of operators and non-closed constructions. The constructions with higher precedence come first. For infix and prefix symbols, we write “*
” to mean “any symbol starting with *”.

Construction or operatorAssociativity
prefix-symbol–
. .( .[ .{ (see section 12.11)–
#
left
function application, constructor application, tag application, assert, lazyleft
- -. (prefix)–
**
 lsl lsr asrright
*
 /
 %
 mod land lor lxorleft
+
 -
left
::right
@
 ^
right
=
 <
 >
 |
 &
 $
 !=left
& &&right
or ||right
,–
<- :=right
if–
;right
let match fun function try–

It is simple to test or refresh one’s understanding:

# 3 + 3 mod 2, 3 + (3 mod 2), (3 + 3) mod 2;;
- : int * int * int = (4, 4, 0)

11.7.2 Basic expressions

Constants

An expression consisting in a constant evaluates to this constant. For example, 3.14 or [||].

Value paths

An expression consisting in an access path evaluates to the value bound to this path in the current evaluation environment. The path can be either a value name or an access path to a value component of a module.

# Float.ArrayLabels.to_list;;
- : Float.ArrayLabels.t -> float list = <fun>

Parenthesized expressions

The expressions ( expr ) and begin expr end have the same value as expr. The two constructs are semantically equivalent, but it is good style to use begin 
 end inside control structures:

        if 
 then begin 
 ; 
 end else begin 
 ; 
 end

and ( 
 ) for the other grouping situations.

# let x = 1 + 2 * 3 let y = (1 + 2) * 3;;
val x : int = 7 val y : int = 9
# let f a b = if a = b then print_endline "Equal" else begin print_string "Not Equal: "; print_int a; print_string " and "; print_int b; print_newline () end;;
val f : int -> int -> unit = <fun>

Parenthesized expressions can contain a type constraint, as in ( expr : typexpr ). This constraint forces the type of expr to be compatible with typexpr.

Parenthesized expressions can also contain coercions ( expr [: typexpr] :> typexpr) (see subsection 11.7.7 below).

Function application

Function application is denoted by juxtaposition of (possibly labeled) expressions. The expression expr argument1 
 argumentn evaluates the expression expr and those appearing in argument1 to argumentn. The expression expr must evaluate to a functional value f, which is then applied to the values of argument1, 
, argumentn.

The order in which the expressions expr, argument1, 
, argumentn are evaluated is not specified.

# List.fold_left ( + ) 0 [1; 2; 3; 4; 5];;
- : int = 15

Arguments and parameters are matched according to their respective labels. Argument order is irrelevant, except among arguments with the same label, or no label.

# ListLabels.fold_left ~f:( @ ) ~init:[] [[1; 2; 3]; [4; 5; 6]; [7; 8; 9]];;
- : int list = [1; 2; 3; 4; 5; 6; 7; 8; 9]

If a parameter is specified as optional (label prefixed by ?) in the type of expr, the corresponding argument will be automatically wrapped with the constructor Some, except if the argument itself is also prefixed by ?, in which case it is passed as is.

# let fullname ?title first second = match title with | Some t -> t ^ " " ^ first ^ " " ^ second | None -> first ^ " " ^ second let name = fullname ~title:"Mrs" "Jane" "Fisher" let address ?title first second town = fullname ?title first second ^ "\n" ^ town;;
val fullname : ?title:string -> string -> string -> string = <fun> val name : string = "Mrs Jane Fisher" val address : ?title:string -> string -> string -> string -> string = <fun>

If a non-labeled argument is passed, and its corresponding parameter is preceded by one or several optional parameters, then these parameters are defaulted, i.e. the value None will be passed for them. All other missing parameters (without corresponding argument), both optional and non-optional, will be kept, and the result of the function will still be a function of these missing parameters to the body of f.

# let fullname ?title first second = match title with | Some t -> t ^ " " ^ first ^ " " ^ second | None -> first ^ " " ^ second let name = fullname "Jane" "Fisher";;
val fullname : ?title:string -> string -> string -> string = <fun> val name : string = "Jane Fisher"

In all cases but exact match of order and labels, without optional parameters, the function type should be known at the application point. This can be ensured by adding a type constraint. Principality of the derivation can be checked in the -principal mode.

As a special case, OCaml supports labels-omitted full applications: if the function has a known arity, all the arguments are unlabeled, and their number matches the number of non-optional parameters, then labels are ignored and non-optional parameters are matched in their definition order. Optional arguments are defaulted. This omission of labels is discouraged and results in a warning, see 13.5.1.

Function definition

Two syntactic forms are provided to define functions. The first form is introduced by the keyword function:

functionpattern1->expr1 
|
 
|patternn->exprn

This expression evaluates to a functional value with one argument. When this function is applied to a value v, this value is matched against each pattern pattern1 to patternn. If one of these matchings succeeds, that is, if the value v matches the pattern patterni for some i, then the expression expri associated to the selected pattern is evaluated, and its value becomes the value of the function application. The evaluation of expri takes place in an environment enriched by the bindings performed during the matching.

If several patterns match the argument v, the one that occurs first in the function definition is selected. If none of the patterns matches the argument, the exception Match_failure is raised.

# (function (0, 0) -> "both zero" | (0, _) -> "first only zero" | (_, 0) -> "second only zero" | (_, _) -> "neither zero") (7, 0);;
- : string = "second only zero"

The other form of function definition is introduced by the keyword fun:

fun parameter1 
 parametern -> expr

This expression is equivalent to:

fun parameter1 -> 
 fun parametern -> expr
# let f = (fun a -> fun b -> fun c -> a + b + c) let g = (fun a b c -> a + b + c);;
val f : int -> int -> int -> int = <fun> val g : int -> int -> int -> int = <fun>

An optional type constraint typexpr can be added before -> to enforce the type of the result to be compatible with the constraint typexpr:

fun parameter1 
 parametern : typexpr -> expr

is equivalent to

fun parameter1 -> 
 fun parametern -> (expr : typexpr )

Beware of the small syntactic difference between a type constraint on the last parameter

fun parameter1 
 (parametern:typexpr)-> expr

and one on the result

fun parameter1 
 parametern: typexpr -> expr
# let eq = fun (a : int) (b : int) -> a = b let eq2 = fun a b : bool -> a = b let eq3 = fun (a : int) (b : int) : bool -> a = b;;
val eq : int -> int -> bool = <fun> val eq2 : 'a -> 'a -> bool = <fun> val eq3 : int -> int -> bool = <fun>

The parameter patterns ~lab and ~(lab [: typ]) are shorthands for respectively ~lab:lab and ~lab:(lab [: typ]), and similarly for their optional counterparts.

# let bool_map ~cmp:(cmp : int -> int -> bool) l = List.map cmp l let bool_map' ~(cmp : int -> int -> bool) l = List.map cmp l;;
val bool_map : cmp:(int -> int -> bool) -> int list -> (int -> bool) list = <fun> val bool_map' : cmp:(int -> int -> bool) -> int list -> (int -> bool) list = <fun>

A function of the form fun ? lab :( pattern = expr0 ) -> expr is equivalent to

fun ? lab : ident -> let pattern = match ident with Some ident -> ident | None -> expr0 in expr

where ident is a fresh variable, except that it is unspecified when expr0 is evaluated.

# let open_file_for_input ?binary filename = match binary with | Some true -> open_in_bin filename | Some false | None -> open_in filename let open_file_for_input' ?(binary=false) filename = if binary then open_in_bin filename else open_in filename;;
val open_file_for_input : ?binary:bool -> string -> in_channel = <fun> val open_file_for_input' : ?binary:bool -> string -> in_channel = <fun>

After these two transformations, expressions are of the form

fun [label1] pattern1 -> 
 fun [labeln] patternn -> expr

If we ignore labels, which will only be meaningful at function application, this is equivalent to

function pattern1 -> 
 function patternn -> expr

That is, the fun expression above evaluates to a curried function with n arguments: after applying this function n times to the values v1 
 vn, the values will be matched in parallel against the patterns pattern1 
 patternn. If the matching succeeds, the function returns the value of expr in an environment enriched by the bindings performed during the matchings. If the matching fails, the exception Match_failure is raised.

Guards in pattern-matchings

The cases of a pattern matching (in the function, match and try constructs) can include guard expressions, which are arbitrary boolean expressions that must evaluate to true for the match case to be selected. Guards occur just before the -> token and are introduced by the when keyword:

functionpattern1   [when   cond1]->expr1 
|
 
|patternn    [when   condn]->exprn

Matching proceeds as described before, except that if the value matches some pattern patterni which has a guard condi, then the expression condi is evaluated (in an environment enriched by the bindings performed during matching). If condi evaluates to true, then expri is evaluated and its value returned as the result of the matching, as usual. But if condi evaluates to false, the matching is resumed against the patterns following patterni.

# let rec repeat f = function | 0 -> () | n when n > 0 -> f (); repeat f (n - 1) | _ -> raise (Invalid_argument "repeat");;
val repeat : (unit -> 'a) -> int -> unit = <fun>

Local definitions

The let and let rec constructs bind value names locally. The construct

let pattern1 = expr1 and 
 and patternn = exprn in expr

evaluates expr1 
 exprn in some unspecified order and matches their values against the patterns pattern1 
 patternn. If the matchings succeed, expr is evaluated in the environment enriched by the bindings performed during matching, and the value of expr is returned as the value of the whole let expression. If one of the matchings fails, the exception Match_failure is raised.

# let v = let x = 1 in [x; x; x] let v' = let a, b = (1, 2) in a + b let v'' = let a = 1 and b = 2 in a + b;;
val v : int list = [1; 1; 1] val v' : int = 3 val v'' : int = 3

An alternate syntax is provided to bind variables to functional values: instead of writing

let ident = fun parameter1 
 parameterm -> expr

in a let expression, one may instead write

let ident parameter1 
 parameterm = expr
# let f = fun x -> fun y -> fun z -> x + y + z let f' = fun x y z -> x + y + z let f'' x y z = x + y + z;;
val f : int -> int -> int -> int = <fun> val f' : int -> int -> int -> int = <fun> val f'' : int -> int -> int -> int = <fun>

Recursive definitions of names are introduced by let rec:

let rec pattern1 = expr1 and 
 and patternn = exprn in expr

The only difference with the let construct described above is that the bindings of names to values performed by the pattern-matching are considered already performed when the expressions expr1 to exprn are evaluated. That is, the expressions expr1 to exprn can reference identifiers that are bound by one of the patterns pattern1, 
, patternn, and expect them to have the same value as in expr, the body of the let rec construct.

# let rec even = function 0 -> true | n -> odd (n - 1) and odd = function 0 -> false | n -> even (n - 1) in even 1000;;
- : bool = true

The recursive definition is guaranteed to behave as described above if the expressions expr1 to exprn are function definitions (fun 
 or function 
), and the patterns pattern1 
 patternn are just value names, as in:

let rec name1 = fun 
 and 
 and namen = fun 
 in expr

This defines name1 
 namen as mutually recursive functions local to expr.

The behavior of other forms of let rec definitions is implementation-dependent. The current implementation also supports a certain class of recursive definitions of non-functional values, as explained in section 12.1.

Local exceptions

(Introduced in OCaml 4.04)

It is possible to define local exceptions in expressions: let exception constr-decl in expr .

# let map_empty_on_negative f l = let exception Negative in let aux x = if x < 0 then raise Negative else f x in try List.map aux l with Negative -> [];;
val map_empty_on_negative : (int -> 'a) -> int list -> 'a list = <fun>

The syntactic scope of the exception constructor is the inner expression, but nothing prevents exception values created with this constructor from escaping this scope. Two executions of the definition above result in two incompatible exception constructors (as for any exception definition). For instance:

# let gen () = let exception A in A let () = assert(gen () = gen ());;
Exception: Assert_failure ("expr.etex", 3, 9).

Explicit polymorphic type annotations

(Introduced in OCaml 3.12)

Polymorphic type annotations in let-definitions behave in a way similar to polymorphic methods:

let pattern1 : typ1 
 typn . typexpr = expr

These annotations explicitly require the defined value to be polymorphic, and allow one to use this polymorphism in recursive occurrences (when using let rec). Note however that this is a normal polymorphic type, unifiable with any instance of itself.

11.7.3 Control structures

Sequence

The expression expr1 ; expr2 evaluates expr1 first, then expr2, and returns the value of expr2.

# let print_pair (a, b) = print_string "("; print_string (string_of_int a); print_string ","; print_string (string_of_int b); print_endline ")";;
val print_pair : int * int -> unit = <fun>

Conditional

The expression if expr1 then expr2 else expr3 evaluates to the value of expr2 if expr1 evaluates to the boolean true, and to the value of expr3 if expr1 evaluates to the boolean false.

# let rec factorial x = if x <= 1 then 1 else x * factorial (x - 1);;
val factorial : int -> int = <fun>

The else expr3 part can be omitted, in which case it defaults to else ().

# let debug = ref false let log msg = if !debug then prerr_endline msg;;
val debug : bool ref = {contents = false} val log : string -> unit = <fun>

Case expression

The expression

matchexpr 
withpattern1->expr1 
|
 
|patternn->exprn

matches the value of expr against the patterns pattern1 to patternn. If the matching against patterni succeeds, the associated expression expri is evaluated, and its value becomes the value of the whole match expression. The evaluation of expri takes place in an environment enriched by the bindings performed during matching. If several patterns match the value of expr, the one that occurs first in the match expression is selected.

# let rec sum l = match l with | [] -> 0 | h :: t -> h + sum t;;
val sum : int list -> int = <fun>

If none of the patterns match the value of expr, the exception Match_failure is raised.

# let unoption o = match o with | Some x -> x ;;
Warning 8 [partial-match]: this pattern-matching is not exhaustive. Here is an example of a case that is not matched: None val unoption : 'a option -> 'a = <fun>
# let l = List.map unoption [Some 1; Some 10; None; Some 2];;
Exception: Match_failure ("expr.etex", 2, 2).

Boolean operators

The expression expr1 && expr2 evaluates to true if both expr1 and expr2 evaluate to true; otherwise, it evaluates to false. The first component, expr1, is evaluated first. The second component, expr2, is not evaluated if the first component evaluates to false. Hence, the expression expr1 && expr2 behaves exactly as

if expr1 then expr2 else false.

The expression expr1 || expr2 evaluates to true if one of the expressions expr1 and expr2 evaluates to true; otherwise, it evaluates to false. The first component, expr1, is evaluated first. The second component, expr2, is not evaluated if the first component evaluates to true. Hence, the expression expr1 || expr2 behaves exactly as

if expr1 then true else expr2.

The boolean operators & and or are deprecated synonyms for (respectively) && and ||.

# let xor a b = (a || b) && not (a && b);;
val xor : bool -> bool -> bool = <fun>

Loops

The expression while expr1 do expr2 done repeatedly evaluates expr2 while expr1 evaluates to true. The loop condition expr1 is evaluated and tested at the beginning of each iteration. The whole while 
 done expression evaluates to the unit value ().

# let chars_of_string s = let i = ref 0 in let chars = ref [] in while !i < String.length s do chars := s.[!i] :: !chars; i := !i + 1 done; List.rev !chars;;
val chars_of_string : string -> char list = <fun>

The expression for name = expr1 to expr2 do expr3 done first evaluates the expressions expr1 and expr2 (the boundaries) into integer values n and p. Then, the loop body expr3 is repeatedly evaluated in an environment where name is successively bound to the values n, n+1, 
, p−1, p. The loop body is never evaluated if n > p.

# let chars_of_string s = let l = ref [] in for p = 0 to String.length s - 1 do l := s.[p] :: !l done; List.rev !l;;
val chars_of_string : string -> char list = <fun>

The expression for name = expr1 downto expr2 do expr3 done evaluates similarly, except that name is successively bound to the values n, n−1, 
, p+1, p. The loop body is never evaluated if n < p.

# let chars_of_string s = let l = ref [] in for p = String.length s - 1 downto 0 do l := s.[p] :: !l done; !l;;
val chars_of_string : string -> char list = <fun>

In both cases, the whole for expression evaluates to the unit value ().

Exception handling

The expression

try expr 
withpattern1->expr1 
|
 
|patternn->exprn

evaluates the expression expr and returns its value if the evaluation of expr does not raise any exception. If the evaluation of expr raises an exception, the exception value is matched against the patterns pattern1 to patternn. If the matching against patterni succeeds, the associated expression expri is evaluated, and its value becomes the value of the whole try expression. The evaluation of expri takes place in an environment enriched by the bindings performed during matching. If several patterns match the value of expr, the one that occurs first in the try expression is selected. If none of the patterns matches the value of expr, the exception value is raised again, thereby transparently “passing through” the try construct.

# let find_opt p l = try Some (List.find p l) with Not_found -> None;;
val find_opt : ('a -> bool) -> 'a list -> 'a option = <fun>

11.7.4 Operations on data structures

Products

The expression expr1 , 
 , exprn evaluates to the n-tuple of the values of expressions expr1 to exprn. The evaluation order of the subexpressions is not specified.

# (1 + 2 * 3, (1 + 2) * 3, 1 + (2 * 3));;
- : int * int * int = (7, 9, 7)

Variants

The expression constr expr evaluates to the unary variant value whose constructor is constr, and whose argument is the value of expr. Similarly, the expression constr ( expr1 , 
 , exprn ) evaluates to the n-ary variant value whose constructor is constr and whose arguments are the values of expr1, 
, exprn.

The expression constr (expr1, 
, exprn) evaluates to the variant value whose constructor is constr, and whose arguments are the values of expr1 
 exprn.

# type t = Var of string | Not of t | And of t * t | Or of t * t let test = And (Var "x", Not (Or (Var "y", Var "z")));;
type t = Var of string | Not of t | And of t * t | Or of t * t val test : t = And (Var "x", Not (Or (Var "y", Var "z")))

For lists, some syntactic sugar is provided. The expression expr1 :: expr2 stands for the constructor ( :: ) applied to the arguments ( expr1 , expr2 ), and therefore evaluates to the list whose head is the value of expr1 and whose tail is the value of expr2. The expression [ expr1 ; 
 ; exprn ] is equivalent to expr1 :: 
 :: exprn :: [], and therefore evaluates to the list whose elements are the values of expr1 to exprn.

# 0 :: [1; 2; 3] = 0 :: 1 :: 2 :: 3 :: [];;
- : bool = true

Polymorphic variants

The expression `tag-name expr evaluates to the polymorphic variant value whose tag is tag-name, and whose argument is the value of expr.

# let with_counter x = `V (x, ref 0);;
val with_counter : 'a -> [> `V of 'a * int ref ] = <fun>

Records

The expression { field1 [= expr1] ; 
 ; fieldn [= exprn ]} evaluates to the record value { field1 = v1; 
; fieldn = vn } where vi is the value of expri for i = 1,
 , n. A single identifier fieldk stands for fieldk = fieldk, and a qualified identifier module-path . fieldk stands for module-path . fieldk = fieldk. The fields field1 to fieldn must all belong to the same record type; each field of this record type must appear exactly once in the record expression, though they can appear in any order. The order in which expr1 to exprn are evaluated is not specified. Optional type constraints can be added after each field { field1 : typexpr1 = expr1 ;
 ; fieldn : typexprn = exprn } to force the type of fieldk to be compatible with typexprk.

# type t = {house_no : int; street : string; town : string; postcode : string} let address x = Printf.sprintf "The occupier\n%i %s\n%s\n%s" x.house_no x.street x.town x.postcode;;
type t = { house_no : int; street : string; town : string; postcode : string; } val address : t -> string = <fun>

The expression { expr with field1 [= expr1] ; 
 ; fieldn [= exprn] } builds a fresh record with fields field1 
 fieldn equal to expr1 
 exprn, and all other fields having the same value as in the record expr. In other terms, it returns a shallow copy of the record expr, except for the fields field1 
 fieldn, which are initialized to expr1 
 exprn. As previously, single identifier fieldk stands for fieldk = fieldk, a qualified identifier module-path . fieldk stands for module-path . fieldk = fieldk and it is possible to add an optional type constraint on each field being updated with { expr with field1 : typexpr1 = expr1 ; 
 ; fieldn : typexprn = exprn }.

# type t = {house_no : int; street : string; town : string; postcode : string} let uppercase_town address = {address with town = String.uppercase_ascii address.town};;
type t = { house_no : int; street : string; town : string; postcode : string; } val uppercase_town : t -> t = <fun>

The expression expr1 . field evaluates expr1 to a record value, and returns the value associated to field in this record value.

The expression expr1 . field <- expr2 evaluates expr1 to a record value, which is then modified in-place by replacing the value associated to field in this record by the value of expr2. This operation is permitted only if field has been declared mutable in the definition of the record type. The whole expression expr1 . field <- expr2 evaluates to the unit value ().

# type t = {mutable upper : int; mutable lower : int; mutable other : int} let stats = {upper = 0; lower = 0; other = 0} let collect = String.iter (function | 'A'..'Z' -> stats.upper <- stats.upper + 1 | 'a'..'z' -> stats.lower <- stats.lower + 1 | _ -> stats.other <- stats.other + 1);;
type t = { mutable upper : int; mutable lower : int; mutable other : int; } val stats : t = {upper = 0; lower = 0; other = 0} val collect : string -> unit = <fun>

Arrays

The expression [| expr1 ; 
 ; exprn |] evaluates to a n-element array, whose elements are initialized with the values of expr1 to exprn respectively. The order in which these expressions are evaluated is unspecified.

The expression expr1 .( expr2 ) returns the value of element number expr2 in the array denoted by expr1. The first element has number 0; the last element has number n−1, where n is the size of the array. The exception Invalid_argument is raised if the access is out of bounds.

The expression expr1 .( expr2 ) <- expr3 modifies in-place the array denoted by expr1, replacing element number expr2 by the value of expr3. The exception Invalid_argument is raised if the access is out of bounds. The value of the whole expression is ().

# let scale arr n = for x = 0 to Array.length arr - 1 do arr.(x) <- arr.(x) * n done let x = [|1; 10; 100|] let _ = scale x 2;;
val scale : int array -> int -> unit = <fun> val x : int array = [|2; 20; 200|]

Strings

The expression expr1 .[ expr2 ] returns the value of character number expr2 in the string denoted by expr1. The first character has number 0; the last character has number n−1, where n is the length of the string. The exception Invalid_argument is raised if the access is out of bounds.

# let iter f s = for x = 0 to String.length s - 1 do f s.[x] done;;
val iter : (char -> 'a) -> string -> unit = <fun>

The expression expr1 .[ expr2 ] <- expr3 modifies in-place the string denoted by expr1, replacing character number expr2 by the value of expr3. The exception Invalid_argument is raised if the access is out of bounds. The value of the whole expression is (). Note: this possibility is offered only for backward compatibility with older versions of OCaml and will be removed in a future version. New code should use byte sequences and the Bytes.set function.

11.7.5 Operators

Symbols from the class infix-symbol, as well as the keywords *, +, -, -., =, !=, <, >, or, ||, &, &&, :=, mod, land, lor, lxor, lsl, lsr, and asr can appear in infix position (between two expressions). Symbols from the class prefix-symbol, as well as the keywords - and -. can appear in prefix position (in front of an expression).

# (( * ), ( := ), ( || ));;
- : (int -> int -> int) * ('a ref -> 'a -> unit) * (bool -> bool -> bool) = (<fun>, <fun>, <fun>)

Infix and prefix symbols do not have a fixed meaning: they are simply interpreted as applications of functions bound to the names corresponding to the symbols. The expression prefix-symbol expr is interpreted as the application ( prefix-symbol ) expr. Similarly, the expression expr1 infix-symbol expr2 is interpreted as the application ( infix-symbol ) expr1 expr2.

The table below lists the symbols defined in the initial environment and their initial meaning. (See the description of the core library module Stdlib in chapter 27 for more details). Their meaning may be changed at any time using let ( infix-op ) name1 name2 = 


# let ( + ), ( - ), ( * ), ( / ) = Int64.(add, sub, mul, div);;
val ( + ) : int64 -> int64 -> int64 = <fun> val ( - ) : int64 -> int64 -> int64 = <fun> val ( * ) : int64 -> int64 -> int64 = <fun> val ( / ) : int64 -> int64 -> int64 = <fun>

Note: the operators &&, ||, and ~- are handled specially and it is not advisable to change their meaning.

The keywords - and -. can appear both as infix and prefix operators. When they appear as prefix operators, they are interpreted respectively as the functions (~-) and (~-.).

OperatorInitial meaning
+Integer addition.
- (infix)Integer subtraction.
~- - (prefix)Integer negation.
*Integer multiplication.
/Integer division. Raise Division_by_zero if second argument is zero.
modInteger modulus. Raise Division_by_zero if second argument is zero.
landBitwise logical “and” on integers.
lorBitwise logical “or” on integers.
lxorBitwise logical “exclusive or” on integers.
lslBitwise logical shift left on integers.
lsrBitwise logical shift right on integers.
asrBitwise arithmetic shift right on integers.
+.Floating-point addition.
-. (infix)Floating-point subtraction.
~-. -. (prefix)Floating-point negation.
*.Floating-point multiplication.
/.Floating-point division.
**Floating-point exponentiation.
@ List concatenation.
^ String concatenation.
! Dereferencing (return the current contents of a reference).
:=Reference assignment (update the reference given as first argument with the value of the second argument).
= Structural equality test.
<> Structural inequality test.
== Physical equality test.
!= Physical inequality test.
< Test “less than”.
<= Test “less than or equal”.
> Test “greater than”.
>= Test “greater than or equal”.
&& &Boolean conjunction.
|| orBoolean disjunction.

11.7.6 Objects

Object creation

When class-path evaluates to a class body, new class-path evaluates to a new object containing the instance variables and methods of this class.

# class of_list (lst : int list) = object val mutable l = lst method next = match l with | [] -> raise (Failure "empty list"); | h::t -> l <- t; h end let a = new of_list [1; 1; 2; 3; 5; 8; 13] let b = new of_list;;
class of_list : int list -> object val mutable l : int list method next : int end val a : of_list = <obj> val b : int list -> of_list = <fun>

When class-path evaluates to a class function, new class-path evaluates to a function expecting the same number of arguments and returning a new object of this class.

Immediate object creation

Creating directly an object through the object class-body end construct is operationally equivalent to defining locally a class class-name = object class-body end —see sections 11.9.2 and following for the syntax of class-body— and immediately creating a single object from it by new class-name.

# let o = object val secret = 99 val password = "unlock" method get guess = if guess <> password then None else Some secret end;;
val o : < get : string -> int option > = <obj>

The typing of immediate objects is slightly different from explicitly defining a class in two respects. First, the inferred object type may contain free type variables. Second, since the class body of an immediate object will never be extended, its self type can be unified with a closed object type.

Method invocation

The expression expr # method-name invokes the method method-name of the object denoted by expr.

# class of_list (lst : int list) = object val mutable l = lst method next = match l with | [] -> raise (Failure "empty list"); | h::t -> l <- t; h end let a = new of_list [1; 1; 2; 3; 5; 8; 13] let third = ignore a#next; ignore a#next; a#next;;
class of_list : int list -> object val mutable l : int list method next : int end val a : of_list = <obj> val third : int = 2

If method-name is a polymorphic method, its type should be known at the invocation site. This is true for instance if expr is the name of a fresh object (let ident = new class-path 
 ) or if there is a type constraint. Principality of the derivation can be checked in the -principal mode.

Accessing and modifying instance variables

The instance variables of a class are visible only in the body of the methods defined in the same class or a class that inherits from the class defining the instance variables. The expression inst-var-name evaluates to the value of the given instance variable. The expression inst-var-name <- expr assigns the value of expr to the instance variable inst-var-name, which must be mutable. The whole expression inst-var-name <- expr evaluates to ().

# class of_list (lst : int list) = object val mutable l = lst method next = match l with (* access instance variable *) | [] -> raise (Failure "empty list"); | h::t -> l <- t; h (* modify instance variable *) end;;
class of_list : int list -> object val mutable l : int list method next : int end

Object duplication

An object can be duplicated using the library function Oo.copy (see module Oo). Inside a method, the expression {< [inst-var-name [= expr] { ; inst-var-name [= expr] }] >} returns a copy of self with the given instance variables replaced by the values of the associated expressions. A single instance variable name id stands for id = id. Other instance variables have the same value in the returned object as in self.

# let o = object val secret = 99 val password = "unlock" method get guess = if guess <> password then None else Some secret method with_new_secret s = {< secret = s >} end;;
val o : < get : string -> int option; with_new_secret : int -> 'a > as 'a = <obj>

11.7.7 Coercions

Expressions whose type contains object or polymorphic variant types can be explicitly coerced (weakened) to a supertype. The expression (expr :> typexpr) coerces the expression expr to type typexpr. The expression (expr : typexpr1 :> typexpr2) coerces the expression expr from type typexpr1 to type typexpr2.

The former operator will sometimes fail to coerce an expression expr from a type typ1 to a type typ2 even if type typ1 is a subtype of type typ2: in the current implementation it only expands two levels of type abbreviations containing objects and/or polymorphic variants, keeping only recursion when it is explicit in the class type (for objects). As an exception to the above algorithm, if both the inferred type of expr and typ are ground (i.e. do not contain type variables), the former operator behaves as the latter one, taking the inferred type of expr as typ1. In case of failure with the former operator, the latter one should be used.

It is only possible to coerce an expression expr from type typ1 to type typ2, if the type of expr is an instance of typ1 (like for a type annotation), and typ1 is a subtype of typ2. The type of the coerced expression is an instance of typ2. If the types contain variables, they may be instantiated by the subtyping algorithm, but this is only done after determining whether typ1 is a potential subtype of typ2. This means that typing may fail during this latter unification step, even if some instance of typ1 is a subtype of some instance of typ2. In the following paragraphs we describe the subtyping relation used.

Object types

A fixed object type admits as subtype any object type that includes all its methods. The types of the methods shall be subtypes of those in the supertype. Namely,

< met1 : typ1 ; 
 ; metn : typn >

is a supertype of

< met1 : typâ€Č1 ; 
 ; metn : typâ€Čn ; metn+1 : typâ€Čn+1 ; 
 ; metn+m : typâ€Čn+m  [; ..] >

which may contain an ellipsis .. if every typi is a supertype of the corresponding typâ€Či.

A monomorphic method type can be a supertype of a polymorphic method type. Namely, if typ is an instance of typâ€Č, then 'a1 
 'an . typâ€Č is a subtype of typ.

Inside a class definition, newly defined types are not available for subtyping, as the type abbreviations are not yet completely defined. There is an exception for coercing self to the (exact) type of its class: this is allowed if the type of self does not appear in a contravariant position in the class type, i.e. if there are no binary methods.

Polymorphic variant types

A polymorphic variant type typ is a subtype of another polymorphic variant type typâ€Č if the upper bound of typ (i.e. the maximum set of constructors that may appear in an instance of typ) is included in the lower bound of typâ€Č, and the types of arguments for the constructors of typ are subtypes of those in typâ€Č. Namely,

[[<] `C1 of typ1 | 
 | `Cn of typn ]

which may be a shrinkable type, is a subtype of

[[>] `C1 of typâ€Č1 | 
 | `Cn of typâ€Čn | `Cn+1 of typâ€Čn+1 | 
 | `Cn+m of typâ€Čn+m ]

which may be an extensible type, if every typi is a subtype of typâ€Či.

Variance

Other types do not introduce new subtyping, but they may propagate the subtyping of their arguments. For instance, typ1 * typ2 is a subtype of typâ€Č1 * typâ€Č2 when typ1 and typ2 are respectively subtypes of typâ€Č1 and typâ€Č2. For function types, the relation is more subtle: typ1 -> typ2 is a subtype of typâ€Č1 -> typâ€Č2 if typ1 is a supertype of typâ€Č1 and typ2 is a subtype of typâ€Č2. For this reason, function types are covariant in their second argument (like tuples), but contravariant in their first argument. Mutable types, like array or ref are neither covariant nor contravariant, they are nonvariant, that is they do not propagate subtyping.

For user-defined types, the variance is automatically inferred: a parameter is covariant if it has only covariant occurrences, contravariant if it has only contravariant occurrences, variance-free if it has no occurrences, and nonvariant otherwise. A variance-free parameter may change freely through subtyping, it does not have to be a subtype or a supertype. For abstract and private types, the variance must be given explicitly (see section 11.8.1), otherwise the default is nonvariant. This is also the case for constrained arguments in type definitions.

11.7.8 Other

Assertion checking

OCaml supports the assert construct to check debugging assertions. The expression assert expr evaluates the expression expr and returns () if expr evaluates to true. If it evaluates to false the exception Assert_failure is raised with the source file name and the location of expr as arguments. Assertion checking can be turned off with the -noassert compiler option. In this case, expr is not evaluated at all.

# let f a b c = assert (a <= b && b <= c); (b -. a) /. (c -. b);;
val f : float -> float -> float -> float = <fun>

As a special case, assert false is reduced to raise (Assert_failure ...), which gives it a polymorphic type. This means that it can be used in place of any expression (for example as a branch of any pattern-matching). It also means that the assert false “assertions” cannot be turned off by the -noassert option.

# let min_known_nonempty = function | [] -> assert false | l -> List.hd (List.sort compare l);;
val min_known_nonempty : 'a list -> 'a = <fun>

Lazy expressions

The expression lazy expr returns a value v of type Lazy.t that encapsulates the computation of expr. The argument expr is not evaluated at this point in the program. Instead, its evaluation will be performed the first time the function Lazy.force is applied to the value v, returning the actual value of expr. Subsequent applications of Lazy.force to v do not evaluate expr again. Applications of Lazy.force may be implicit through pattern matching (see 11.6).

# let lazy_greeter = lazy (print_string "Hello, World!\n");;
val lazy_greeter : unit lazy_t = <lazy>
# Lazy.force lazy_greeter;;
Hello, World! - : unit = ()

Local modules

The expression let module module-name = module-expr in expr locally binds the module expression module-expr to the identifier module-name during the evaluation of the expression expr. It then returns the value of expr. For example:

# let remove_duplicates comparison_fun string_list = let module StringSet = Set.Make(struct type t = string let compare = comparison_fun end) in StringSet.elements (List.fold_right StringSet.add string_list StringSet.empty);;
val remove_duplicates : (string -> string -> int) -> string list -> string list = <fun>

Local opens

The expressions let open module-path in expr and module-path.(expr) are strictly equivalent. These constructions locally open the module referred to by the module path module-path in the respective scope of the expression expr.

# let map_3d_matrix f m = let open Array in map (map (map f)) m let map_3d_matrix' f = Array.(map (map (map f)));;
val map_3d_matrix : ('a -> 'b) -> 'a array array array -> 'b array array array = <fun> val map_3d_matrix' : ('a -> 'b) -> 'a array array array -> 'b array array array = <fun>

When the body of a local open expression is delimited by [ ], [| |], or { }, the parentheses can be omitted. For expression, parentheses can also be omitted for {< >}. For example, module-path.[expr] is equivalent to module-path.([expr]), and module-path.[| expr |] is equivalent to module-path.([| expr |]).

# let vector = Random.[|int 255; int 255; int 255; int 255|];;
val vector : int array = [|220; 90; 247; 144|]

11.8 Type and exception definitions

11.8.1 Type definitions

Type definitions bind type constructors to data types: either variant types, record types, type abbreviations, or abstract data types. They also bind the value constructors and record fields associated with the definition.

type-definition::= type [nonrec] typedef { and typedef }
 
typedef::= [type-params] typeconstr-name type-information
 
type-information::= [type-equation] [type-representation] { type-constraint }
 
type-equation::= = typexpr
 
type-representation::= = [|] constr-decl { | constr-decl }
 âˆŁâ€„= record-decl
 âˆŁâ€„= |
 
type-params::= type-param
 âˆŁâ€„( type-param { , type-param } )
 
type-param::= [ext-variance] ' ident
 
ext-variance::= variance [injectivity]
 âˆŁâ€„injectivity [variance]
 
variance::= +
 âˆŁâ€„-
 
injectivity::= !
 
record-decl::= { field-decl { ; field-decl } [;] }
 
constr-decl::= (constr-name ∣ [] ∣ (::)) [ of constr-args ]
 
constr-args::= typexpr { * typexpr }
 
field-decl::= [mutable] field-name : poly-typexpr
 
type-constraint::= constraint typexpr = typexpr

See also the following language extensions: private types, generalized algebraic datatypes, attributes, extension nodes, extensible variant types and inline records.

Type definitions are introduced by the type keyword, and consist in one or several simple definitions, possibly mutually recursive, separated by the and keyword. Each simple definition defines one type constructor.

A simple definition consists in a lowercase identifier, possibly preceded by one or several type parameters, and followed by an optional type equation, then an optional type representation, and then a constraint clause. The identifier is the name of the type constructor being defined.

type colour =
  | Red | Green | Blue | Yellow | Black | White
  | RGB of {r : int; g : int; b : int}

type 'a tree = Lf | Br of 'a * 'a tree * 'a;;

type t = {decoration : string; substance : t'}
and t' = Int of int | List of t list

In the right-hand side of type definitions, references to one of the type constructor name being defined are considered as recursive, unless type is followed by nonrec. The nonrec keyword was introduced in OCaml 4.02.2.

The optional type parameters are either one type variable ' ident, for type constructors with one parameter, or a list of type variables ('ident1,
,'identn), for type constructors with several parameters. Each type parameter may be prefixed by a variance constraint + (resp. -) indicating that the parameter is covariant (resp. contravariant), and an injectivity annotation ! indicating that the parameter can be deduced from the whole type. These type parameters can appear in the type expressions of the right-hand side of the definition, optionally restricted by a variance constraint ; i.e. a covariant parameter may only appear on the right side of a functional arrow (more precisely, follow the left branch of an even number of arrows), and a contravariant parameter only the left side (left branch of an odd number of arrows). If the type has a representation or an equation, and the parameter is free (i.e. not bound via a type constraint to a constructed type), its variance constraint is checked but subtyping etc. will use the inferred variance of the parameter, which may be less restrictive; otherwise (i.e. for abstract types or non-free parameters), the variance must be given explicitly, and the parameter is invariant if no variance is given.

The optional type equation = typexpr makes the defined type equivalent to the type expression typexpr: one can be substituted for the other during typing. If no type equation is given, a new type is generated: the defined type is incompatible with any other type.

The optional type representation describes the data structure representing the defined type, by giving the list of associated constructors (if it is a variant type) or associated fields (if it is a record type). If no type representation is given, nothing is assumed on the structure of the type besides what is stated in the optional type equation.

The type representation = [|] constr-decl { | constr-decl } describes a variant type. The constructor declarations constr-decl1, 
, constr-decln describe the constructors associated to this variant type. The constructor declaration constr-name of typexpr1 * 
 * typexprn declares the name constr-name as a non-constant constructor, whose arguments have types typexpr1 
typexprn. The constructor declaration constr-name declares the name constr-name as a constant constructor. Constructor names must be capitalized.

The type representation = { field-decl { ; field-decl } [;] } describes a record type. The field declarations field-decl1, 
, field-decln describe the fields associated to this record type. The field declaration field-name : poly-typexpr declares field-name as a field whose argument has type poly-typexpr. The field declaration mutable field-name : poly-typexpr behaves similarly; in addition, it allows physical modification of this field. Immutable fields are covariant, mutable fields are non-variant. Both mutable and immutable fields may have explicitly polymorphic types. The polymorphism of the contents is statically checked whenever a record value is created or modified. Extracted values may have their types instantiated.

The two components of a type definition, the optional equation and the optional representation, can be combined independently, giving rise to four typical situations:

Abstract type: no equation, no representation.
 
When appearing in a module signature, this definition specifies nothing on the type constructor, besides its number of parameters: its representation is hidden and it is assumed incompatible with any other type.
Type abbreviation: an equation, no representation.
 
This defines the type constructor as an abbreviation for the type expression on the right of the = sign.
New variant type or record type: no equation, a representation.
 
This generates a new type constructor and defines associated constructors or fields, through which values of that type can be directly built or inspected.
Re-exported variant type or record type: an equation, a representation.
 
In this case, the type constructor is defined as an abbreviation for the type expression given in the equation, but in addition the constructors or fields given in the representation remain attached to the defined type constructor. The type expression in the equation part must agree with the representation: it must be of the same kind (record or variant) and have exactly the same constructors or fields, in the same order, with the same arguments. Moreover, the new type constructor must have the same arity and the same type constraints as the original type constructor.

The type variables appearing as type parameters can optionally be prefixed by + or - to indicate that the type constructor is covariant or contravariant with respect to this parameter. This variance information is used to decide subtyping relations when checking the validity of :> coercions (see section 11.7.7).

For instance, type +'a t declares t as an abstract type that is covariant in its parameter; this means that if the type τ is a subtype of the type σ, then τ t is a subtype of σ t. Similarly, type -'a t declares that the abstract type t is contravariant in its parameter: if τ is a subtype of σ, then σ t is a subtype of τ t. If no + or - variance annotation is given, the type constructor is assumed non-variant in the corresponding parameter. For instance, the abstract type declaration type 'a t means that τ t is neither a subtype nor a supertype of σ t if τ is subtype of σ.

The variance indicated by the + and - annotations on parameters is enforced only for abstract and private types, or when there are type constraints. Otherwise, for abbreviations, variant and record types without type constraints, the variance properties of the type constructor are inferred from its definition, and the variance annotations are only checked for conformance with the definition.

Injectivity annotations are only necessary for abstract types and private row types, since they can otherwise be deduced from the type declaration: all parameters are injective for record and variant type declarations (including extensible types); for type abbreviations a parameter is injective if it has an injective occurrence in its defining equation (be it private or not). For constrained type parameters in type abbreviations, they are injective if either they appear at an injective position in the body, or if all their type variables are injective; in particular, if a constrained type parameter contains a variable that doesn’t appear in the body, it cannot be injective.

The construct constraint ' ident = typexpr allows the specification of type parameters. Any actual type argument corresponding to the type parameter ident has to be an instance of typexpr (more precisely, ident and typexpr are unified). Type variables of typexpr can appear in the type equation and the type declaration.

11.8.2 Exception definitions

exception-definition::= exception constr-decl
 âˆŁâ€„exception constr-name = constr

Exception definitions add new constructors to the built-in variant type exn of exception values. The constructors are declared as for a definition of a variant type.

# exception E of int * string;;
exception E of int * string

The form exception constr-decl generates a new exception, distinct from all other exceptions in the system. The form exception constr-name = constr gives an alternate name to an existing exception.

# exception E of int * string exception F = E let eq = E (1, "one") = F (1, "one");;
exception E of int * string exception F of int * string val eq : bool = true

11.9 Classes

Classes are defined using a small language, similar to the module language.

11.9.1 Class types

Class types are the class-level equivalent of type expressions: they specify the general shape and type properties of classes.

class-type::= [[?]label-name:] typexpr -> class-type
 âˆŁâ€„  class-body-type
 
class-body-type::= object [( typexpr )] { class-field-spec } end
 âˆŁâ€„ [[ typexpr { , typexpr } ]] classtype-path
 âˆŁâ€„ let open module-path in class-body-type
 
class-field-spec::= inherit class-body-type
 âˆŁâ€„ val [mutable] [virtual] inst-var-name : typexpr
 âˆŁâ€„ val virtual mutable inst-var-name : typexpr
 âˆŁâ€„ method [private] [virtual] method-name : poly-typexpr
 âˆŁâ€„ method virtual private method-name : poly-typexpr
 âˆŁâ€„ constraint typexpr = typexpr

See also the following language extensions: attributes and extension nodes.

Simple class expressions

The expression classtype-path is equivalent to the class type bound to the name classtype-path. Similarly, the expression [ typexpr1 , 
 typexprn ] classtype-path is equivalent to the parametric class type bound to the name classtype-path, in which type parameters have been instantiated to respectively typexpr1, 
typexprn.

Class function type

The class type expression typexpr -> class-type is the type of class functions (functions from values to classes) that take as argument a value of type typexpr and return as result a class of type class-type.

Class body type

The class type expression object [( typexpr )] { class-field-spec } end is the type of a class body. It specifies its instance variables and methods. In this type, typexpr is matched against the self type, therefore providing a name for the self type.

A class body will match a class body type if it provides definitions for all the components specified in the class body type, and these definitions meet the type requirements given in the class body type. Furthermore, all methods either virtual or public present in the class body must also be present in the class body type (on the other hand, some instance variables and concrete private methods may be omitted). A virtual method will match a concrete method, which makes it possible to forget its implementation. An immutable instance variable will match a mutable instance variable.

Local opens

Local opens are supported in class types since OCaml 4.06.

Inheritance

The inheritance construct inherit class-body-type provides for inclusion of methods and instance variables from other class types. The instance variable and method types from class-body-type are added into the current class type.

Instance variable specification

A specification of an instance variable is written val [mutable] [virtual] inst-var-name : typexpr, where inst-var-name is the name of the instance variable and typexpr its expected type. The flag mutable indicates whether this instance variable can be physically modified. The flag virtual indicates that this instance variable is not initialized. It can be initialized later through inheritance.

An instance variable specification will hide any previous specification of an instance variable of the same name.

Method specification

The specification of a method is written method [private] method-name : poly-typexpr, where method-name is the name of the method and poly-typexpr its expected type, possibly polymorphic. The flag private indicates that the method cannot be accessed from outside the object.

The polymorphism may be left implicit in public method specifications: any type variable which is not bound to a class parameter and does not appear elsewhere inside the class specification will be assumed to be universal, and made polymorphic in the resulting method type. Writing an explicit polymorphic type will disable this behaviour.

If several specifications are present for the same method, they must have compatible types. Any non-private specification of a method forces it to be public.

Virtual method specification

A virtual method specification is written method [private] virtual method-name : poly-typexpr, where method-name is the name of the method and poly-typexpr its expected type.

Constraints on type parameters

The construct constraint typexpr1 = typexpr2 forces the two type expressions to be equal. This is typically used to specify type parameters: in this way, they can be bound to specific type expressions.

11.9.2 Class expressions

Class expressions are the class-level equivalent of value expressions: they evaluate to classes, thus providing implementations for the specifications expressed in class types.

class-expr::= class-path
 âˆŁâ€„ [ typexpr { , typexpr } ] class-path
 âˆŁâ€„ ( class-expr )
 âˆŁâ€„ ( class-expr : class-type )
 âˆŁâ€„ class-expr { argument }+
 âˆŁâ€„ fun { parameter }+ -> class-expr
 âˆŁâ€„ let [rec] let-binding { and let-binding } in class-expr
 âˆŁâ€„ object class-body end
 âˆŁâ€„ let open module-path in class-expr
 
class-field::= inherit class-expr [as lowercase-ident]
 âˆŁâ€„ inherit! class-expr [as lowercase-ident]
 âˆŁâ€„ val [mutable] inst-var-name [: typexpr] = expr
 âˆŁâ€„ val! [mutable] inst-var-name [: typexpr] = expr
 âˆŁâ€„ val [mutable] virtual inst-var-name : typexpr
 âˆŁâ€„ val virtual mutable inst-var-name : typexpr
 âˆŁâ€„ method [private] method-name { parameter } [: typexpr] = expr
 âˆŁâ€„ method! [private] method-name { parameter } [: typexpr] = expr
 âˆŁâ€„ method [private] method-name : poly-typexpr = expr
 âˆŁâ€„ method! [private] method-name : poly-typexpr = expr
 âˆŁâ€„ method [private] virtual method-name : poly-typexpr
 âˆŁâ€„ method virtual private method-name : poly-typexpr
 âˆŁâ€„ constraint typexpr = typexpr
 âˆŁâ€„ initializer expr

See also the following language extensions: locally abstract types, attributes and extension nodes.

Simple class expressions

The expression class-path evaluates to the class bound to the name class-path. Similarly, the expression [ typexpr1 , 
 typexprn ] class-path evaluates to the parametric class bound to the name class-path, in which type parameters have been instantiated respectively to typexpr1, 
typexprn.

The expression ( class-expr ) evaluates to the same module as class-expr.

The expression ( class-expr : class-type ) checks that class-type matches the type of class-expr (that is, that the implementation class-expr meets the type specification class-type). The whole expression evaluates to the same class as class-expr, except that all components not specified in class-type are hidden and can no longer be accessed.

Class application

Class application is denoted by juxtaposition of (possibly labeled) expressions. It denotes the class whose constructor is the first expression applied to the given arguments. The arguments are evaluated as for expression application, but the constructor itself will only be evaluated when objects are created. In particular, side-effects caused by the application of the constructor will only occur at object creation time.

Class function

The expression fun [[?]label-name:]pattern -> class-expr evaluates to a function from values to classes. When this function is applied to a value v, this value is matched against the pattern pattern and the result is the result of the evaluation of class-expr in the extended environment.

Conversion from functions with default values to functions with patterns only works identically for class functions as for normal functions.

The expression

fun parameter1 
 parametern -> class-expr

is a short form for

fun parameter1 -> 
 fun parametern -> expr

Local definitions

The let and let rec constructs bind value names locally, as for the core language expressions.

If a local definition occurs at the very beginning of a class definition, it will be evaluated when the class is created (just as if the definition was outside of the class). Otherwise, it will be evaluated when the object constructor is called.

Local opens

Local opens are supported in class expressions since OCaml 4.06.

Class body

class-body::=  [( pattern [: typexpr] )] { class-field }

The expression object class-body end denotes a class body. This is the prototype for an object : it lists the instance variables and methods of an object of this class.

A class body is a class value: it is not evaluated at once. Rather, its components are evaluated each time an object is created.

In a class body, the pattern ( pattern [: typexpr] ) is matched against self, therefore providing a binding for self and self type. Self can only be used in method and initializers.

Self type cannot be a closed object type, so that the class remains extensible.

Since OCaml 4.01, it is an error if the same method or instance variable name is defined several times in the same class body.

Inheritance

The inheritance construct inherit class-expr allows reusing methods and instance variables from other classes. The class expression class-expr must evaluate to a class body. The instance variables, methods and initializers from this class body are added into the current class. The addition of a method will override any previously defined method of the same name.

An ancestor can be bound by appending as lowercase-ident to the inheritance construct. lowercase-ident is not a true variable and can only be used to select a method, i.e. in an expression lowercase-ident # method-name. This gives access to the method method-name as it was defined in the parent class even if it is redefined in the current class. The scope of this ancestor binding is limited to the current class. The ancestor method may be called from a subclass but only indirectly.

Instance variable definition

The definition val [mutable] inst-var-name = expr adds an instance variable inst-var-name whose initial value is the value of expression expr. The flag mutable allows physical modification of this variable by methods.

An instance variable can only be used in the methods and initializers that follow its definition.

Since version 3.10, redefinitions of a visible instance variable with the same name do not create a new variable, but are merged, using the last value for initialization. They must have identical types and mutability. However, if an instance variable is hidden by omitting it from an interface, it will be kept distinct from other instance variables with the same name.

Virtual instance variable definition

A variable specification is written val [mutable] virtual inst-var-name : typexpr. It specifies whether the variable is modifiable, and gives its type.

Virtual instance variables were added in version 3.10.

Method definition

A method definition is written method method-name = expr. The definition of a method overrides any previous definition of this method. The method will be public (that is, not private) if any of the definition states so.

A private method, method private method-name = expr, is a method that can only be invoked on self (from other methods of the same object, defined in this class or one of its subclasses). This invocation is performed using the expression value-name # method-name, where value-name is directly bound to self at the beginning of the class definition. Private methods do not appear in object types. A method may have both public and private definitions, but as soon as there is a public one, all subsequent definitions will be made public.

Methods may have an explicitly polymorphic type, allowing them to be used polymorphically in programs (even for the same object). The explicit declaration may be done in one of three ways: (1) by giving an explicit polymorphic type in the method definition, immediately after the method name, i.e. method [private] method-name : { ' ident }+ . typexpr = expr; (2) by a forward declaration of the explicit polymorphic type through a virtual method definition; (3) by importing such a declaration through inheritance and/or constraining the type of self.

Some special expressions are available in method bodies for manipulating instance variables and duplicating self:

expr::= 

 âˆŁâ€„inst-var-name <- expr
 âˆŁâ€„{< [ inst-var-name = expr { ; inst-var-name = expr } [;] ] >}

The expression inst-var-name <- expr modifies in-place the current object by replacing the value associated to inst-var-name by the value of expr. Of course, this instance variable must have been declared mutable.

The expression {< inst-var-name1 = expr1 ; 
 ; inst-var-namen = exprn >} evaluates to a copy of the current object in which the values of instance variables inst-var-name1, 
, inst-var-namen have been replaced by the values of the corresponding expressions expr1, 
, exprn.

Virtual method definition

A method specification is written method [private] virtual method-name : poly-typexpr. It specifies whether the method is public or private, and gives its type. If the method is intended to be polymorphic, the type must be explicitly polymorphic.

Explicit overriding

Since Ocaml 3.12, the keywords inherit!, val! and method! have the same semantics as inherit, val and method, but they additionally require the definition they introduce to be overriding. Namely, method! requires method-name to be already defined in this class, val! requires inst-var-name to be already defined in this class, and inherit! requires class-expr to override some definitions. If no such overriding occurs, an error is signaled.

As a side-effect, these 3 keywords avoid the warnings 7 (method override) and 13 (instance variable override). Note that warning 7 is disabled by default.

Constraints on type parameters

The construct constraint typexpr1 = typexpr2 forces the two type expressions to be equals. This is typically used to specify type parameters: in that way they can be bound to specific type expressions.

Initializers

A class initializer initializer expr specifies an expression that will be evaluated whenever an object is created from the class, once all its instance variables have been initialized.

11.9.3 Class definitions

class-definition::= class class-binding { and class-binding }
 
class-binding::= [virtual] [[ type-parameters ]] class-name { parameter } [: class-type]  = class-expr
 
type-parameters::= ' ident { , ' ident }

A class definition class class-binding { and class-binding } is recursive. Each class-binding defines a class-name that can be used in the whole expression except for inheritance. It can also be used for inheritance, but only in the definitions that follow its own.

A class binding binds the class name class-name to the value of expression class-expr. It also binds the class type class-name to the type of the class, and defines two type abbreviations : class-name and # class-name. The first one is the type of objects of this class, while the second is more general as it unifies with the type of any object belonging to a subclass (see section 11.4).

Virtual class

A class must be flagged virtual if one of its methods is virtual (that is, appears in the class type, but is not actually defined). Objects cannot be created from a virtual class.

Type parameters

The class type parameters correspond to the ones of the class type and of the two type abbreviations defined by the class binding. They must be bound to actual types in the class definition using type constraints. So that the abbreviations are well-formed, type variables of the inferred type of the class must either be type parameters or be bound in the constraint clause.

11.9.4 Class specifications

class-specification::= class class-spec { and class-spec }
 
class-spec::= [virtual] [[ type-parameters ]] class-name : class-type

This is the counterpart in signatures of class definitions. A class specification matches a class definition if they have the same type parameters and their types match.

11.9.5 Class type definitions

classtype-definition::= class type classtype-def { and classtype-def }
 
classtype-def::= [virtual] [[ type-parameters ]] class-name = class-body-type

A class type definition class class-name = class-body-type defines an abbreviation class-name for the class body type class-body-type. As for class definitions, two type abbreviations class-name and # class-name are also defined. The definition can be parameterized by some type parameters. If any method in the class type body is virtual, the definition must be flagged virtual.

Two class type definitions match if they have the same type parameters and they expand to matching types.

11.10 Module types (module specifications)

Module types are the module-level equivalent of type expressions: they specify the general shape and type properties of modules.

module-type::= modtype-path
 âˆŁâ€„sig { specification [;;] } end
 âˆŁâ€„functor ( module-name : module-type ) -> module-type
 âˆŁâ€„module-type -> module-type
 âˆŁâ€„module-type with mod-constraint { and mod-constraint }
 âˆŁâ€„( module-type )
 
mod-constraint::= type [type-params] typeconstr type-equation { type-constraint }
 âˆŁâ€„module module-path = extended-module-path
 
specification::= val value-name : typexpr
 âˆŁâ€„external value-name : typexpr = external-declaration
 âˆŁâ€„type-definition
 âˆŁâ€„exception constr-decl
 âˆŁâ€„class-specification
 âˆŁâ€„classtype-definition
 âˆŁâ€„module module-name : module-type
 âˆŁâ€„module module-name { ( module-name : module-type ) } : module-type
 âˆŁâ€„module type modtype-name
 âˆŁâ€„module type modtype-name = module-type
 âˆŁâ€„open module-path
 âˆŁâ€„include module-type

See also the following language extensions: recovering the type of a module, substitution inside a signature, type-level module aliases, attributes, extension nodes, generative functors, and module type substitutions.

11.10.1 Simple module types

The expression modtype-path is equivalent to the module type bound to the name modtype-path. The expression ( module-type ) denotes the same type as module-type.

11.10.2 Signatures

Signatures are type specifications for structures. Signatures sig 
 end are collections of type specifications for value names, type names, exceptions, module names and module type names. A structure will match a signature if the structure provides definitions (implementations) for all the names specified in the signature (and possibly more), and these definitions meet the type requirements given in the signature.

An optional ;; is allowed after each specification in a signature. It serves as a syntactic separator with no semantic meaning.

Value specifications

A specification of a value component in a signature is written val value-name : typexpr, where value-name is the name of the value and typexpr its expected type.

The form external value-name : typexpr = external-declaration is similar, except that it requires in addition the name to be implemented as the external function specified in external-declaration (see chapter 22).

Type specifications

A specification of one or several type components in a signature is written type typedef { and typedef } and consists of a sequence of mutually recursive definitions of type names.

Each type definition in the signature specifies an optional type equation = typexpr and an optional type representation = constr-decl 
 or = { field-decl 
 }. The implementation of the type name in a matching structure must be compatible with the type expression specified in the equation (if given), and have the specified representation (if given). Conversely, users of that signature will be able to rely on the type equation or type representation, if given. More precisely, we have the following four situations:

Abstract type: no equation, no representation.
 
Names that are defined as abstract types in a signature can be implemented in a matching structure by any kind of type definition (provided it has the same number of type parameters). The exact implementation of the type will be hidden to the users of the structure. In particular, if the type is implemented as a variant type or record type, the associated constructors and fields will not be accessible to the users; if the type is implemented as an abbreviation, the type equality between the type name and the right-hand side of the abbreviation will be hidden from the users of the structure. Users of the structure consider that type as incompatible with any other type: a fresh type has been generated.
Type abbreviation: an equation = typexpr, no representation.
 
The type name must be implemented by a type compatible with typexpr. All users of the structure know that the type name is compatible with typexpr.
New variant type or record type: no equation, a representation.
 
The type name must be implemented by a variant type or record type with exactly the constructors or fields specified. All users of the structure have access to the constructors or fields, and can use them to create or inspect values of that type. However, users of the structure consider that type as incompatible with any other type: a fresh type has been generated.
Re-exported variant type or record type: an equation, a representation.
 
This case combines the previous two: the representation of the type is made visible to all users, and no fresh type is generated.

Exception specification

The specification exception constr-decl in a signature requires the matching structure to provide an exception with the name and arguments specified in the definition, and makes the exception available to all users of the structure.

Class specifications

A specification of one or several classes in a signature is written class class-spec { and class-spec } and consists of a sequence of mutually recursive definitions of class names.

Class specifications are described more precisely in section 11.9.4.

Class type specifications

A specification of one or several class types in a signature is written class type classtype-def { and classtype-def } and consists of a sequence of mutually recursive definitions of class type names. Class type specifications are described more precisely in section 11.9.5.

Module specifications

A specification of a module component in a signature is written module module-name : module-type, where module-name is the name of the module component and module-type its expected type. Modules can be nested arbitrarily; in particular, functors can appear as components of structures and functor types as components of signatures.

For specifying a module component that is a functor, one may write

module module-name ( name1 : module-type1 ) 
 ( namen : module-typen ) : module-type

instead of

module module-name : functor ( name1 : module-type1 ) -> 
 -> module-type

Module type specifications

A module type component of a signature can be specified either as a manifest module type or as an abstract module type.

An abstract module type specification module type modtype-name allows the name modtype-name to be implemented by any module type in a matching signature, but hides the implementation of the module type to all users of the signature.

A manifest module type specification module type modtype-name = module-type requires the name modtype-name to be implemented by the module type module-type in a matching signature, but makes the equality between modtype-name and module-type apparent to all users of the signature.

Opening a module path

The expression open module-path in a signature does not specify any components. It simply affects the parsing of the following items of the signature, allowing components of the module denoted by module-path to be referred to by their simple names name instead of path accesses module-path . name. The scope of the open stops at the end of the signature expression.

Including a signature

The expression include module-type in a signature performs textual inclusion of the components of the signature denoted by module-type. It behaves as if the components of the included signature were copied at the location of the include. The module-type argument must refer to a module type that is a signature, not a functor type.

11.10.3 Functor types

The module type expression functor ( module-name : module-type1 ) -> module-type2 is the type of functors (functions from modules to modules) that take as argument a module of type module-type1 and return as result a module of type module-type2. The module type module-type2 can use the name module-name to refer to type components of the actual argument of the functor. If the type module-type2 does not depend on type components of module-name, the module type expression can be simplified with the alternative short syntax module-type1 -> module-type2 . No restrictions are placed on the type of the functor argument; in particular, a functor may take another functor as argument (“higher-order” functor).

When the result module type is itself a functor,

functor ( name1 : module-type1 ) -> 
 -> functor ( namen : module-typen ) -> module-type

one may use the abbreviated form

functor ( name1 : module-type1 ) 
 ( namen : module-typen ) -> module-type

11.10.4 The with operator

Assuming module-type denotes a signature, the expression module-type with mod-constraint { and mod-constraint } denotes the same signature where type equations have been added to some of the type specifications, as described by the constraints following the with keyword. The constraint type [type-parameters] typeconstr = typexpr adds the type equation = typexpr to the specification of the type component named typeconstr of the constrained signature. The constraint module module-path = extended-module-path adds type equations to all type components of the sub-structure denoted by module-path, making them equivalent to the corresponding type components of the structure denoted by extended-module-path.

For instance, if the module type name S is bound to the signature

        sig type t module M: (sig type u end) end

then S with type t=int denotes the signature

        sig type t=int module M: (sig type u end) end

and S with module M = N denotes the signature

        sig type t module M: (sig type u=N.u end) end

A functor taking two arguments of type S that share their t component is written

        functor (A: S) (B: S with type t = A.t) ...

Constraints are added left to right. After each constraint has been applied, the resulting signature must be a subtype of the signature before the constraint was applied. Thus, the with operator can only add information on the type components of a signature, but never remove information.

11.11 Module expressions (module implementations)

Module expressions are the module-level equivalent of value expressions: they evaluate to modules, thus providing implementations for the specifications expressed in module types.

module-expr::= module-path
 âˆŁâ€„struct [ module-items ] end
 âˆŁâ€„functor ( module-name : module-type ) -> module-expr
 âˆŁâ€„module-expr ( module-expr )
 âˆŁâ€„( module-expr )
 âˆŁâ€„( module-expr : module-type )
 
module-items::= { ;; } ( definition ∣ expr ) { { ;; } ( definition ∣ ;; expr) } { ;; }
 
definition::= let [rec] let-binding { and let-binding }
 âˆŁâ€„external value-name : typexpr = external-declaration
 âˆŁâ€„type-definition
 âˆŁâ€„exception-definition
 âˆŁâ€„class-definition
 âˆŁâ€„classtype-definition
 âˆŁâ€„module module-name { ( module-name : module-type ) } [ : module-type ]  = module-expr
 âˆŁâ€„module type modtype-name = module-type
 âˆŁâ€„open module-path
 âˆŁâ€„include module-expr

See also the following language extensions: recursive modules, first-class modules, overriding in open statements, attributes, extension nodes and generative functors.

11.11.1 Simple module expressions

The expression module-path evaluates to the module bound to the name module-path.

The expression ( module-expr ) evaluates to the same module as module-expr.

The expression ( module-expr : module-type ) checks that the type of module-expr is a subtype of module-type, that is, that all components specified in module-type are implemented in module-expr, and their implementation meets the requirements given in module-type. In other terms, it checks that the implementation module-expr meets the type specification module-type. The whole expression evaluates to the same module as module-expr, except that all components not specified in module-type are hidden and can no longer be accessed.

11.11.2 Structures

Structures struct 
 end are collections of definitions for value names, type names, exceptions, module names and module type names. The definitions are evaluated in the order in which they appear in the structure. The scopes of the bindings performed by the definitions extend to the end of the structure. As a consequence, a definition may refer to names bound by earlier definitions in the same structure.

For compatibility with toplevel phrases (chapter 14), optional ;; are allowed after and before each definition in a structure. These ;; have no semantic meanings. Similarly, an expr preceded by ;; is allowed as a component of a structure. It is equivalent to let _ = expr, i.e. expr is evaluated for its side-effects but is not bound to any identifier. If expr is the first component of a structure, the preceding ;; can be omitted.

Value definitions

A value definition let [rec] let-binding { and let-binding } bind value names in the same way as a let 
 in 
 expression (see section 11.7.2). The value names appearing in the left-hand sides of the bindings are bound to the corresponding values in the right-hand sides.

A value definition external value-name : typexpr = external-declaration implements value-name as the external function specified in external-declaration (see chapter 22).

Type definitions

A definition of one or several type components is written type typedef { and typedef } and consists of a sequence of mutually recursive definitions of type names.

Exception definitions

Exceptions are defined with the syntax exception constr-decl or exception constr-name = constr.

Class definitions

A definition of one or several classes is written class class-binding { and class-binding } and consists of a sequence of mutually recursive definitions of class names. Class definitions are described more precisely in section 11.9.3.

Class type definitions

A definition of one or several classes is written class type classtype-def { and classtype-def } and consists of a sequence of mutually recursive definitions of class type names. Class type definitions are described more precisely in section 11.9.5.

Module definitions

The basic form for defining a module component is module module-name = module-expr, which evaluates module-expr and binds the result to the name module-name.

One can write

module module-name : module-type = module-expr

instead of

module module-name = ( module-expr : module-type ).

Another derived form is

module module-name ( name1 : module-type1 ) 
 ( namen : module-typen ) = module-expr

which is equivalent to

module module-name = functor ( name1 : module-type1 ) -> 
 -> module-expr

Module type definitions

A definition for a module type is written module type modtype-name = module-type. It binds the name modtype-name to the module type denoted by the expression module-type.

Opening a module path

The expression open module-path in a structure does not define any components nor perform any bindings. It simply affects the parsing of the following items of the structure, allowing components of the module denoted by module-path to be referred to by their simple names name instead of path accesses module-path . name. The scope of the open stops at the end of the structure expression.

Including the components of another structure

The expression include module-expr in a structure re-exports in the current structure all definitions of the structure denoted by module-expr. For instance, if you define a module S as below

module S = struct type t = int let x = 2 end

defining the module B as

module B = struct include S let y = (x + 1 : t) end

is equivalent to defining it as

module B = struct type t = S.t let x = S.x let y = (x + 1 : t) end

The difference between open and include is that open simply provides short names for the components of the opened structure, without defining any components of the current structure, while include also adds definitions for the components of the included structure.

11.11.3 Functors

Functor definition

The expression functor ( module-name : module-type ) -> module-expr evaluates to a functor that takes as argument modules of the type module-type1, binds module-name to these modules, evaluates module-expr in the extended environment, and returns the resulting modules as results. No restrictions are placed on the type of the functor argument; in particular, a functor may take another functor as argument (“higher-order” functor).

When the result module expression is itself a functor,

functor ( name1 : module-type1 ) -> 
 -> functor ( namen : module-typen ) -> module-expr

one may use the abbreviated form

functor ( name1 : module-type1 ) 
 ( namen : module-typen ) -> module-expr

Functor application

The expression module-expr1 ( module-expr2 ) evaluates module-expr1 to a functor and module-expr2 to a module, and applies the former to the latter. The type of module-expr2 must match the type expected for the arguments of the functor module-expr1.

11.12 Compilation units

unit-interface::= { specification [;;] }
 
unit-implementation::= [ module-items ]

Compilation units bridge the module system and the separate compilation system. A compilation unit is composed of two parts: an interface and an implementation. The interface contains a sequence of specifications, just as the inside of a sig 
 end signature expression. The implementation contains a sequence of definitions and expressions, just as the inside of a struct 
 end module expression. A compilation unit also has a name unit-name, derived from the names of the files containing the interface and the implementation (see chapter 13 for more details). A compilation unit behaves roughly as the module definition

module unit-name : sig unit-interface end = struct unit-implementation end

A compilation unit can refer to other compilation units by their names, as if they were regular modules. For instance, if U is a compilation unit that defines a type t, other compilation units can refer to that type under the name U.t; they can also refer to U as a whole structure. Except for names of other compilation units, a unit interface or unit implementation must not have any other free variables. In other terms, the type-checking and compilation of an interface or implementation proceeds in the initial environment

name1 : sig specification1 end 
 namen : sig specificationn end

where name1 
 namen are the names of the other compilation units available in the search path (see chapter 13 for more details) and specification1 
 specificationn are their respective interfaces.

Chapter 12 Language extensions

This chapter describes language extensions and convenience features that are implemented in OCaml, but not described in chapter 11.

12.1 Recursive definitions of values

(Introduced in Objective Caml 1.00)

As mentioned in section 11.7.2, the let rec binding construct, in addition to the definition of recursive functions, also supports a certain class of recursive definitions of non-functional values, such as

let rec name1 = 1 :: name2 and name2 = 2 :: name1 in expr

which binds name1 to the cyclic list 1::2::1::2::
, and name2 to the cyclic list 2::1::2::1::
Informally, the class of accepted definitions consists of those definitions where the defined names occur only inside function bodies or as argument to a data constructor.

More precisely, consider the expression:

let rec name1 = expr1 and 
 and namen = exprn in expr

It will be accepted if each one of expr1 
 exprn is statically constructive with respect to name1 
 namen, is not immediately linked to any of name1 
 namen, and is not an array constructor whose arguments have abstract type.

An expression e is said to be statically constructive with respect to the variables name1 
 namen if at least one of the following conditions is true:

An expression e is said to be immediately linked to the variable name in the following cases:

12.2 Recursive modules

(Introduced in Objective Caml 3.07)

definition::= ...
 âˆŁâ€„module rec module-name : module-type = module-expr  { and module-name : module-type = module-expr }
 
specification::= ...
 âˆŁâ€„module rec module-name : module-type { and module-name: module-type }

Recursive module definitions, introduced by the module rec 
and 
 construction, generalize regular module definitions module module-name = module-expr and module specifications module module-name : module-type by allowing the defining module-expr and the module-type to refer recursively to the module identifiers being defined. A typical example of a recursive module definition is:

module rec A : sig type t = Leaf of string | Node of ASet.t val compare: t -> t -> int end = struct type t = Leaf of string | Node of ASet.t let compare t1 t2 = match (t1, t2) with | (Leaf s1, Leaf s2) -> Stdlib.compare s1 s2 | (Leaf _, Node _) -> 1 | (Node _, Leaf _) -> -1 | (Node n1, Node n2) -> ASet.compare n1 n2 end and ASet : Set.S with type elt = A.t = Set.Make(A)

It can be given the following specification:

module rec A : sig type t = Leaf of string | Node of ASet.t val compare: t -> t -> int end and ASet : Set.S with type elt = A.t

This is an experimental extension of OCaml: the class of recursive definitions accepted, as well as its dynamic semantics are not final and subject to change in future releases.

Currently, the compiler requires that all dependency cycles between the recursively-defined module identifiers go through at least one “safe” module. A module is “safe” if all value definitions that it contains have function types typexpr1 -> typexpr2. Evaluation of a recursive module definition proceeds by building initial values for the safe modules involved, binding all (functional) values to fun _ -> raise Undefined_recursive_module. The defining module expressions are then evaluated, and the initial values for the safe modules are replaced by the values thus computed. If a function component of a safe module is applied during this computation (which corresponds to an ill-founded recursive definition), the Undefined_recursive_module exception is raised at runtime:

module rec M: sig val f: unit -> int end = struct let f () = N.x end and N:sig val x: int end = struct let x = M.f () end
Exception: Undefined_recursive_module ("extensions/recursivemodules.etex", 1, 43).

If there are no safe modules along a dependency cycle, an error is raised

module rec M: sig val x: int end = struct let x = N.y end and N:sig val x: int val y:int end = struct let x = M.x let y = 0 end
Error: Cannot safely evaluate the definition of the following cycle of recursively-defined modules: M -> N -> M. There are no safe modules in this cycle (see manual section 12.2). Module M defines an unsafe value, x . Module N defines an unsafe value, x .

Note that, in the specification case, the module-types must be parenthesized if they use the with mod-constraint construct.

12.3 Private types

Private type declarations in module signatures, of the form type t = private ..., enable libraries to reveal some, but not all aspects of the implementation of a type to clients of the library. In this respect, they strike a middle ground between abstract type declarations, where no information is revealed on the type implementation, and data type definitions and type abbreviations, where all aspects of the type implementation are publicized. Private type declarations come in three flavors: for variant and record types (section 12.3.1), for type abbreviations (section 12.3.2), and for row types (section 12.3.3).

12.3.1 Private variant and record types

(Introduced in Objective Caml 3.07)

type-representation::= ...
 âˆŁâ€„= private [ | ] constr-decl { | constr-decl }
 âˆŁâ€„= private record-decl

Values of a variant or record type declared private can be de-structured normally in pattern-matching or via the expr . field notation for record accesses. However, values of these types cannot be constructed directly by constructor application or record construction. Moreover, assignment on a mutable field of a private record type is not allowed.

The typical use of private types is in the export signature of a module, to ensure that construction of values of the private type always go through the functions provided by the module, while still allowing pattern-matching outside the defining module. For example:

module M : sig type t = private A | B of int val a : t val b : int -> t end = struct type t = A | B of int let a = A let b n = assert (n > 0); B n end

Here, the private declaration ensures that in any value of type M.t, the argument to the B constructor is always a positive integer.

With respect to the variance of their parameters, private types are handled like abstract types. That is, if a private type has parameters, their variance is the one explicitly given by prefixing the parameter by a ‘+’ or a ‘-’, it is invariant otherwise.

12.3.2 Private type abbreviations

(Introduced in Objective Caml 3.11)

type-equation::= ...
 âˆŁâ€„= private typexpr

Unlike a regular type abbreviation, a private type abbreviation declares a type that is distinct from its implementation type typexpr. However, coercions from the type to typexpr are permitted. Moreover, the compiler “knows” the implementation type and can take advantage of this knowledge to perform type-directed optimizations.

The following example uses a private type abbreviation to define a module of nonnegative integers:

module N : sig type t = private int val of_int: int -> t val to_int: t -> int end = struct type t = int let of_int n = assert (n >= 0); n let to_int n = n end

The type N.t is incompatible with int, ensuring that nonnegative integers and regular integers are not confused. However, if x has type N.t, the coercion (x :> int) is legal and returns the underlying integer, just like N.to_int x. Deep coercions are also supported: if l has type N.t list, the coercion (l :> int list) returns the list of underlying integers, like List.map N.to_int l but without copying the list l.

Note that the coercion ( expr :> typexpr ) is actually an abbreviated form, and will only work in presence of private abbreviations if neither the type of expr nor typexpr contain any type variables. If they do, you must use the full form ( expr : typexpr1 :> typexpr2 ) where typexpr1 is the expected type of expr. Concretely, this would be (x : N.t :> int) and (l : N.t list :> int list) for the above examples.

12.3.3 Private row types

(Introduced in Objective Caml 3.09)

type-equation::= ...
 âˆŁâ€„= private typexpr

Private row types are type abbreviations where part of the structure of the type is left abstract. Concretely typexpr in the above should denote either an object type or a polymorphic variant type, with some possibility of refinement left. If the private declaration is used in an interface, the corresponding implementation may either provide a ground instance, or a refined private type.

module M : sig type c = private < x : int; .. > val o : c end = struct class c = object method x = 3 method y = 2 end let o = new c end

This declaration does more than hiding the y method, it also makes the type c incompatible with any other closed object type, meaning that only o will be of type c. In that respect it behaves similarly to private record types. But private row types are more flexible with respect to incremental refinement. This feature can be used in combination with functors.

module F(X : sig type c = private < x : int; .. > end) = struct let get_x (o : X.c) = o#x end module G(X : sig type c = private < x : int; y : int; .. > end) = struct include F(X) let get_y (o : X.c) = o#y end

A polymorphic variant type [t], for example

type t = [ `A of int | `B of bool ]

can be refined in two ways. A definition [u] may add new field to [t], and the declaration

type u = private [> t]

will keep those new fields abstract. Construction of values of type [u] is possible using the known variants of [t], but any pattern-matching will require a default case to handle the potential extra fields. Dually, a declaration [u] may restrict the fields of [t] through abstraction: the declaration

type v = private [< t > `A]

corresponds to private variant types. One cannot create a value of the private type [v], except using the constructors that are explicitly listed as present, (`A n) in this example; yet, when pattern-matching on a [v], one should assume that any of the constructors of [t] could be present.

Similarly to abstract types, the variance of type parameters is not inferred, and must be given explicitly.

12.4 Locally abstract types

(Introduced in OCaml 3.12, short syntax added in 4.03)

parameter::= ...
 âˆŁâ€„( type { typeconstr-name }+ )

The expression fun ( type typeconstr-name ) -> expr introduces a type constructor named typeconstr-name which is considered abstract in the scope of the sub-expression, but then replaced by a fresh type variable. Note that contrary to what the syntax could suggest, the expression fun ( type typeconstr-name ) -> expr itself does not suspend the evaluation of expr as a regular abstraction would. The syntax has been chosen to fit nicely in the context of function declarations, where it is generally used. It is possible to freely mix regular function parameters with pseudo type parameters, as in:

let f = fun (type t) (foo : t list) -> 


and even use the alternative syntax for declaring functions:

let f (type t) (foo : t list) = 


If several locally abstract types need to be introduced, it is possible to use the syntax fun ( type typeconstr-name1 
 typeconstr-namen ) -> expr as syntactic sugar for fun ( type typeconstr-name1 ) -> 
 -> fun ( type typeconstr-namen ) -> expr. For instance,

let f = fun (type t u v) -> fun (foo : (t * u * v) list) -> 
 let f' (type t u v) (foo : (t * u * v) list) = 


This construction is useful because the type constructors it introduces can be used in places where a type variable is not allowed. For instance, one can use it to define an exception in a local module within a polymorphic function.

let f (type t) () = let module M = struct exception E of t end in (fun x -> M.E x), (function M.E x -> Some x | _ -> None)

Here is another example:

let sort_uniq (type s) (cmp : s -> s -> int) = let module S = Set.Make(struct type t = s let compare = cmp end) in fun l -> S.elements (List.fold_right S.add l S.empty)

It is also extremely useful for first-class modules (see section 12.5) and generalized algebraic datatypes (GADTs: see section 12.10).

Polymorphic syntax

(Introduced in OCaml 4.00)

let-binding::= ...
 âˆŁâ€„value-name : type { typeconstr-name }+ . typexpr = expr
 
class-field::= ...
 âˆŁâ€„method [private] method-name : type { typeconstr-name }+ . typexpr = expr
 âˆŁâ€„method! [private] method-name : type { typeconstr-name }+ . typexpr = expr

The (type typeconstr-name) syntax construction by itself does not make polymorphic the type variable it introduces, but it can be combined with explicit polymorphic annotations where needed. The above rule is provided as syntactic sugar to make this easier:

let rec f : type t1 t2. t1 * t2 list -> t1 = 


is automatically expanded into

let rec f : 't1 't2. 't1 * 't2 list -> 't1 = fun (type t1) (type t2) -> ( 
 : t1 * t2 list -> t1)

This syntax can be very useful when defining recursive functions involving GADTs, see the section 12.10 for a more detailed explanation.

The same feature is provided for method definitions.

12.5 First-class modules

(Introduced in OCaml 3.12; pattern syntax and package type inference introduced in 4.00; structural comparison of package types introduced in 4.02.; fewer parens required starting from 4.05)

typexpr::= ...
 âˆŁâ€„(module package-type)
 
module-expr::= ...
 âˆŁâ€„(val expr [: package-type])
 
expr::= ...
 âˆŁâ€„(module module-expr [: package-type])
 
pattern::= ...
 âˆŁâ€„(module module-name [: package-type])
 
package-type::= modtype-path
 âˆŁâ€„modtype-path with package-constraint { and package-constraint }
 
package-constraint::= type typeconstr = typexpr
 

Modules are typically thought of as static components. This extension makes it possible to pack a module as a first-class value, which can later be dynamically unpacked into a module.

The expression ( module module-expr : package-type ) converts the module (structure or functor) denoted by module expression module-expr to a value of the core language that encapsulates this module. The type of this core language value is ( module package-type ). The package-type annotation can be omitted if it can be inferred from the context.

Conversely, the module expression ( val expr : package-type ) evaluates the core language expression expr to a value, which must have type module package-type, and extracts the module that was encapsulated in this value. Again package-type can be omitted if the type of expr is known. If the module expression is already parenthesized, like the arguments of functors are, no additional parens are needed: Map.Make(val key).

The pattern ( module module-name : package-type ) matches a package with type package-type and binds it to module-name. It is not allowed in toplevel let bindings. Again package-type can be omitted if it can be inferred from the enclosing pattern.

The package-type syntactic class appearing in the ( module package-type ) type expression and in the annotated forms represents a subset of module types. This subset consists of named module types with optional constraints of a limited form: only non-parametrized types can be specified.

For type-checking purposes (and starting from OCaml 4.02), package types are compared using the structural comparison of module types.

In general, the module expression ( val expr : package-type ) cannot be used in the body of a functor, because this could cause unsoundness in conjunction with applicative functors. Since OCaml 4.02, this is relaxed in two ways: if package-type does not contain nominal type declarations (i.e. types that are created with a proper identity), then this expression can be used anywhere, and even if it contains such types it can be used inside the body of a generative functor, described in section 12.15. It can also be used anywhere in the context of a local module binding let module module-name = ( val expr1 : package-type ) in expr2.

Basic example

A typical use of first-class modules is to select at run-time among several implementations of a signature. Each implementation is a structure that we can encapsulate as a first-class module, then store in a data structure such as a hash table:

type picture = 
 module type DEVICE = sig val draw : picture -> unit 
 end let devices : (string, (module DEVICE)) Hashtbl.t = Hashtbl.create 17 module SVG = struct 
 end let _ = Hashtbl.add devices "SVG" (module SVG : DEVICE) module PDF = struct 
 end let _ = Hashtbl.add devices "PDF" (module PDF : DEVICE)

We can then select one implementation based on command-line arguments, for instance:

let parse_cmdline () = 
 module Device = (val (let device_name = parse_cmdline () in try Hashtbl.find devices device_name with Not_found -> Printf.eprintf "Unknown device %s\n" device_name; exit 2) : DEVICE)

Alternatively, the selection can be performed within a function:

let draw_using_device device_name picture = let module Device = (val (Hashtbl.find devices device_name) : DEVICE) in Device.draw picture
Advanced examples

With first-class modules, it is possible to parametrize some code over the implementation of a module without using a functor.

let sort (type s) (module Set : Set.S with type elt = s) l = Set.elements (List.fold_right Set.add l Set.empty)
val sort : (module Set.S with type elt = 's) -> 's list -> 's list = <fun>

To use this function, one can wrap the Set.Make functor:

let make_set (type s) cmp = let module S = Set.Make(struct type t = s let compare = cmp end) in (module S : Set.S with type elt = s)
val make_set : ('s -> 's -> int) -> (module Set.S with type elt = 's) = <fun>

12.6 Recovering the type of a module

(Introduced in OCaml 3.12)

module-type::= ...
 âˆŁâ€„module type of module-expr

The construction module type of module-expr expands to the module type (signature or functor type) inferred for the module expression module-expr. To make this module type reusable in many situations, it is intentionally not strengthened: abstract types and datatypes are not explicitly related with the types of the original module. For the same reason, module aliases in the inferred type are expanded.

A typical use, in conjunction with the signature-level include construct, is to extend the signature of an existing structure. In that case, one wants to keep the types equal to types in the original module. This can done using the following idiom.

module type MYHASH = sig include module type of struct include Hashtbl end val replace: ('a, 'b) t -> 'a -> 'b -> unit end

The signature MYHASH then contains all the fields of the signature of the module Hashtbl (with strengthened type definitions), plus the new field replace. An implementation of this signature can be obtained easily by using the include construct again, but this time at the structure level:

module MyHash : MYHASH = struct include Hashtbl let replace t k v = remove t k; add t k v end

Another application where the absence of strengthening comes handy, is to provide an alternative implementation for an existing module.

module MySet : module type of Set = struct 
 end

This idiom guarantees that Myset is compatible with Set, but allows it to represent sets internally in a different way.

12.7 Substituting inside a signature

12.7.1 Destructive substitutions

(Introduced in OCaml 3.12, generalized in 4.06)

mod-constraint::= ...
 âˆŁâ€„type [type-params] typeconstr-name := typexpr
 âˆŁâ€„module module-path := extended-module-path

A “destructive” substitution (with ... := ...) behaves essentially like normal signature constraints (with ... = ...), but it additionally removes the redefined type or module from the signature.

Prior to OCaml 4.06, there were a number of restrictions: one could only remove types and modules at the outermost level (not inside submodules), and in the case of with type the definition had to be another type constructor with the same type parameters.

A natural application of destructive substitution is merging two signatures sharing a type name.

module type Printable = sig type t val print : Format.formatter -> t -> unit end module type Comparable = sig type t val compare : t -> t -> int end module type PrintableComparable = sig include Printable include Comparable with type t := t end

One can also use this to completely remove a field:

module type S = Comparable with type t := int
module type S = sig val compare : int -> int -> int end

or to rename one:

module type S = sig type u include Comparable with type t := u end
module type S = sig type u val compare : u -> u -> int end

Note that you can also remove manifest types, by substituting with the same type.

module type ComparableInt = Comparable with type t = int ;;
module type ComparableInt = sig type t = int val compare : t -> t -> int end
module type CompareInt = ComparableInt with type t := int
module type CompareInt = sig val compare : int -> int -> int end

12.7.2 Local substitution declarations

(Introduced in OCaml 4.08)

specification::= ...
 âˆŁâ€„type type-subst { and type-subst }
 âˆŁâ€„module module-name := extended-module-path
 âˆŁâ€„module type module-name := module-type
 
type-subst::= [type-params] typeconstr-name := typexpr { type-constraint }

Local substitutions behave like destructive substitutions (with ... := ...) but instead of being applied to a whole signature after the fact, they are introduced during the specification of the signature, and will apply to all the items that follow.

This provides a convenient way to introduce local names for types and modules when defining a signature:

module type S = sig type t module Sub : sig type outer := t type t val to_outer : t -> outer end end
module type S = sig type t module Sub : sig type t val to_outer : t -> t/2 end end

Note that, unlike type declarations, type substitution declarations are not recursive, so substitutions like the following are rejected:

# module type S = sig type 'a poly_list := [ `Cons of 'a * 'a poly_list | `Nil ] end ;;
Error: Unbound type constructor poly_list

12.7.3 Module type substitutions

(Introduced in OCaml 4.13)

mod-constraint::= ...
 âˆŁâ€„module type modtype-path = module-type
 âˆŁâ€„module type modtype-path := module-type

Module type substitution essentially behaves like type substitutions. They are useful to refine an abstract module type in a signature into a concrete module type,

# module type ENDO = sig module type T module F: T -> T end module Endo(X: sig module type T end): ENDO with module type T = X.T = struct module type T = X.T module F(X:T) = X end;;
module type ENDO = sig module type T module F : T -> T end module Endo : functor (X : sig module type T end) -> sig module type T = X.T module F : T -> T end

It is also possible to substitute a concrete module type with an equivalent module types.

module type A = sig type x module type R = sig type a = A of x type b end end module type S = sig type a = A of int type b end module type B = A with type x = int and module type R = S

However, such substitutions are never necessary.

Destructive module type substitution removes the module type substitution from the signature

# module type ENDO' = ENDO with module type T := ENDO;;
module type ENDO' = sig module F : ENDO -> ENDO end

If the right hand side of the substitution is not a path, then the destructive substitution is only valid if the left-hand side of the substitution is never used as the type of a first-class module in the original module type.

module type T = sig module type S val x: (module S) end module type Error = T with module type S := sig end
Error: This `with' constraint S := sig end makes a packed module ill-formed.

12.8 Type-level module aliases

(Introduced in OCaml 4.02)

specification::= ...
 âˆŁâ€„module module-name = module-path

The above specification, inside a signature, only matches a module definition equal to module-path. Conversely, a type-level module alias can be matched by itself, or by any supertype of the type of the module it references.

There are several restrictions on module-path:

  1. it should be of the form M0.M1...Mn (i.e. without functor applications);
  2. inside the body of a functor, M0 should not be one of the functor parameters;
  3. inside a recursive module definition, M0 should not be one of the recursively defined modules.

Such specifications are also inferred. Namely, when P is a path satisfying the above constraints,

module N = P

has type

module N = P

Type-level module aliases are used when checking module path equalities. That is, in a context where module name N is known to be an alias for P, not only these two module paths check as equal, but F(N) and F(P) are also recognized as equal. In the default compilation mode, this is the only difference with the previous approach of module aliases having just the same module type as the module they reference.

When the compiler flag -no-alias-deps is enabled, type-level module aliases are also exploited to avoid introducing dependencies between compilation units. Namely, a module alias referring to a module inside another compilation unit does not introduce a link-time dependency on that compilation unit, as long as it is not dereferenced; it still introduces a compile-time dependency if the interface needs to be read, i.e. if the module is a submodule of the compilation unit, or if some type components are referred to. Additionally, accessing a module alias introduces a link-time dependency on the compilation unit containing the module referenced by the alias, rather than the compilation unit containing the alias. Note that these differences in link-time behavior may be incompatible with the previous behavior, as some compilation units might not be extracted from libraries, and their side-effects ignored.

These weakened dependencies make possible to use module aliases in place of the -pack mechanism. Suppose that you have a library Mylib composed of modules A and B. Using -pack, one would issue the command line

ocamlc -pack a.cmo b.cmo -o mylib.cmo

and as a result obtain a Mylib compilation unit, containing physically A and B as submodules, and with no dependencies on their respective compilation units. Here is a concrete example of a possible alternative approach:

  1. Rename the files containing A and B to Mylib__A and Mylib__B.
  2. Create a packing interface Mylib.ml, containing the following lines.
    module A = Mylib__A
    module B = Mylib__B
    
  3. Compile Mylib.ml using -no-alias-deps, and the other files using -no-alias-deps and -open Mylib (the last one is equivalent to adding the line open! Mylib at the top of each file).
    ocamlc -c -no-alias-deps Mylib.ml
    ocamlc -c -no-alias-deps -open Mylib Mylib__*.mli Mylib__*.ml
    
  4. Finally, create a library containing all the compilation units, and export all the compiled interfaces.
    ocamlc -a Mylib*.cmo -o Mylib.cma
    

This approach lets you access A and B directly inside the library, and as Mylib.A and Mylib.B from outside. It also has the advantage that Mylib is no longer monolithic: if you use Mylib.A, only Mylib__A will be linked in, not Mylib__B.

Note the use of double underscores in Mylib__A and Mylib__B. These were chosen on purpose; the compiler uses the following heuristic when printing paths: given a path Lib__fooBar, if Lib.FooBar exists and is an alias for Lib__fooBar, then the compiler will always display Lib.FooBar instead of Lib__fooBar. This way the long Mylib__ names stay hidden and all the user sees is the nicer dot names. This is how the OCaml standard library is compiled.

12.9 Overriding in open statements

(Introduced in OCaml 4.01)

definition::= ...
 âˆŁâ€„ open! module-path
 
specification::= ...
 âˆŁâ€„ open! module-path
 
expr::= ...
 âˆŁâ€„let open! module-path in expr
 
class-body-type::= ...
 âˆŁâ€„ let open! module-path in class-body-type
 
class-expr::= ...
 âˆŁâ€„ let open! module-path in class-expr
 

Since OCaml 4.01, open statements shadowing an existing identifier (which is later used) trigger the warning 44. Adding a ! character after the open keyword indicates that such a shadowing is intentional and should not trigger the warning.

This is also available (since OCaml 4.06) for local opens in class expressions and class type expressions.

12.10 Generalized algebraic datatypes

Generalized algebraic datatypes, or GADTs, extend usual sum types in two ways: constraints on type parameters may change depending on the value constructor, and some type variables may be existentially quantified. They are described in chapter 7.

(Introduced in OCaml 4.00)

constr-decl::= ...
 âˆŁâ€„constr-name : [ constr-args -> ] typexpr
 
type-param::= ...
 âˆŁâ€„[variance] _

Refutation cases. (Introduced in OCaml 4.03)

matching-case::= pattern [when expr] -> expr
 âˆŁâ€„pattern -> .

Explicit naming of existentials. (Introduced in OCaml 4.13.0)

pattern::= ...
 âˆŁâ€„constr ( type { typeconstr-name }+ ) ( pattern )
 

12.11 Syntax for Bigarray access

(Introduced in Objective Caml 3.00)

expr::= ...
 âˆŁâ€„expr .{ expr { , expr } }
 âˆŁâ€„expr .{ expr { , expr } } <- expr

This extension provides syntactic sugar for getting and setting elements in the arrays provided by the Bigarray module.

The short expressions are translated into calls to functions of the Bigarray module as described in the following table.

expressiontranslation
expr0.{expr1}Bigarray.Array1.get expr0 expr1
expr0.{expr1} <-exprBigarray.Array1.set expr0 expr1 expr
expr0.{expr1, expr2}Bigarray.Array2.get expr0 expr1 expr2
expr0.{expr1, expr2} <-exprBigarray.Array2.set expr0 expr1 expr2 expr
expr0.{expr1, expr2, expr3}Bigarray.Array3.get expr0 expr1 expr2 expr3
expr0.{expr1, expr2, expr3} <-exprBigarray.Array3.set expr0 expr1 expr2 expr3 expr
expr0.{expr1, 
, exprn}Bigarray.Genarray.get expr0 [| expr1, 
 , exprn |]
expr0.{expr1, 
, exprn} <-exprBigarray.Genarray.set expr0 [| expr1, 
 , exprn |] expr

The last two entries are valid for any n > 3.

12.12 Attributes

(Introduced in OCaml 4.02, infix notations for constructs other than expressions added in 4.03)

Attributes are “decorations” of the syntax tree which are mostly ignored by the type-checker but can be used by external tools. An attribute is made of an identifier and a payload, which can be a structure, a type expression (prefixed with :), a signature (prefixed with :) or a pattern (prefixed with ?) optionally followed by a when clause:

attr-id::= lowercase-ident
 âˆŁâ€„ capitalized-ident
 âˆŁâ€„ attr-id . attr-id
 
attr-payload::= [ module-items ]
 âˆŁâ€„ : typexpr
 âˆŁâ€„ : [ specification ]
 âˆŁâ€„ ? pattern [when expr]
 

The first form of attributes is attached with a postfix notation on “algebraic” categories:

attribute::= [@ attr-id attr-payload ]
 
expr::= ...
 âˆŁâ€„expr attribute
 
typexpr::= ...
 âˆŁâ€„typexpr attribute
 
pattern::= ...
 âˆŁâ€„pattern attribute
 
module-expr::= ...
 âˆŁâ€„module-expr attribute
 
module-type::= ...
 âˆŁâ€„module-type attribute
 
class-expr::= ...
 âˆŁâ€„class-expr attribute
 
class-type::= ...
 âˆŁâ€„class-type attribute
 

This form of attributes can also be inserted after the `tag-name in polymorphic variant type expressions (tag-spec-first, tag-spec, tag-spec-full) or after the method-name in method-type.

The same syntactic form is also used to attach attributes to labels and constructors in type declarations:

field-decl::= [mutable] field-name : poly-typexpr { attribute }
 
constr-decl::= (constr-name ∣ ()) [ of constr-args ] { attribute }
 

Note: when a label declaration is followed by a semi-colon, attributes can also be put after the semi-colon (in which case they are merged to those specified before).

The second form of attributes are attached to “blocks” such as type declarations, class fields, etc:

item-attribute::= [@@ attr-id attr-payload ]
 
typedef::= ...
 âˆŁâ€„typedef item-attribute
 
exception-definition::= exception constr-decl
 âˆŁâ€„exception constr-name = constr
 
module-items::= [;;] ( definition ∣ expr { item-attribute } ) { [;;] definition ∣ ;; expr { item-attribute } } [;;]
 
class-binding::= ...
 âˆŁâ€„class-binding item-attribute
 
class-spec::= ...
 âˆŁâ€„class-spec item-attribute
 
classtype-def::= ...
 âˆŁâ€„classtype-def item-attribute
 
definition::= let [rec] let-binding { and let-binding }
 âˆŁâ€„external value-name : typexpr = external-declaration { item-attribute }
 âˆŁâ€„type-definition
 âˆŁâ€„exception-definition { item-attribute }
 âˆŁâ€„class-definition
 âˆŁâ€„classtype-definition
 âˆŁâ€„module module-name { ( module-name : module-type ) } [ : module-type ]  = module-expr { item-attribute }
 âˆŁâ€„module type modtype-name = module-type { item-attribute }
 âˆŁâ€„open module-path { item-attribute }
 âˆŁâ€„include module-expr { item-attribute }
 âˆŁâ€„module rec module-name : module-type =  module-expr { item-attribute }  { and module-name : module-type = module-expr  { item-attribute } }
 
specification::= val value-name : typexpr { item-attribute }
 âˆŁâ€„external value-name : typexpr = external-declaration { item-attribute }
 âˆŁâ€„type-definition
 âˆŁâ€„exception constr-decl { item-attribute }
 âˆŁâ€„class-specification
 âˆŁâ€„classtype-definition
 âˆŁâ€„module module-name : module-type { item-attribute }
 âˆŁâ€„module module-name { ( module-name : module-type ) } : module-type { item-attribute }
 âˆŁâ€„module type modtype-name { item-attribute }
 âˆŁâ€„module type modtype-name = module-type { item-attribute }
 âˆŁâ€„open module-path { item-attribute }
 âˆŁâ€„include module-type { item-attribute }
 
class-field-spec::= ...
 âˆŁâ€„class-field-spec item-attribute
 
class-field::= ...
 âˆŁâ€„class-field item-attribute
 

A third form of attributes appears as stand-alone structure or signature items in the module or class sub-languages. They are not attached to any specific node in the syntax tree:

floating-attribute::= [@@@ attr-id attr-payload ]
 
definition::= ...
 âˆŁâ€„floating-attribute
 
specification::= ...
 âˆŁâ€„floating-attribute
 
class-field-spec::= ...
 âˆŁâ€„floating-attribute
 
class-field::= ...
 âˆŁâ€„floating-attribute
 

(Note: contrary to what the grammar above describes, item-attributes cannot be attached to these floating attributes in class-field-spec and class-field.)

It is also possible to specify attributes using an infix syntax. For instance:

let[@foo] x = 2 in x + 1          === (let x = 2 [@@foo] in x + 1)
begin[@foo][@bar x] ... end       === (begin ... end)[@foo][@bar x]
module[@foo] M = ...              === module M = ... [@@foo]
type[@foo] t = T                  === type t = T [@@foo]
method[@foo] m = ...              === method m = ... [@@foo]

For let, the attributes are applied to each bindings:

let[@foo] x = 2 and y = 3 in x + y === (let x = 2 [@@foo] and y = 3 in x + y)
let[@foo] x = 2
and[@bar] y = 3 in x + y           === (let x = 2 [@@foo] and y = 3 [@@bar] in x + y)

12.12.1 Built-in attributes

Some attributes are understood by the type-checker:

module X = struct [@@@warning "+9"] (* locally enable warning 9 in this structure *) 
 end [@@deprecated "Please use module 'Y' instead."] let x = begin[@warning "+9"] [
] end type t = A | B [@@deprecated "Please use type 's' instead."]
let fires_warning_22 x = assert (x >= 0) [@ppwarning "TODO: remove this later"]
Warning 22 [preprocessor]: TODO: remove this later
let rec is_a_tail_call = function | [] -> () | _ :: q -> (is_a_tail_call[@tailcall]) q let rec not_a_tail_call = function | [] -> [] | x :: q -> x :: (not_a_tail_call[@tailcall]) q
Warning 51 [wrong-tailcall-expectation]: expected tailcall
let f x = x [@@inline] let () = (f[@inlined]) ()
type fragile = | Int of int [@warn_on_literal_pattern] | String of string [@warn_on_literal_pattern]
let fragile_match_1 = function | Int 0 -> () | _ -> ()
Warning 52 [fragile-literal-pattern]: Code should not depend on the actual values of this constructor's arguments. They are only for information and may change in future versions. (see manual section 13.5.3) val fragile_match_1 : fragile -> unit = <fun>
let fragile_match_2 = function | String "constant" -> () | _ -> ()
Warning 52 [fragile-literal-pattern]: Code should not depend on the actual values of this constructor's arguments. They are only for information and may change in future versions. (see manual section 13.5.3) val fragile_match_2 : fragile -> unit = <fun>
module Immediate: sig type t [@@immediate] val x: t ref end = struct type t = A | B let x = ref A end
module Int_or_int64 : sig type t [@@immediate64] val zero : t val one : t val add : t -> t -> t end = struct include Sys.Immediate64.Make(Int)(Int64) module type S = sig val zero : t val one : t val add : t -> t -> t end let impl : (module S) = match repr with | Immediate -> (module Int : S) | Non_immediate -> (module Int64 : S) include (val impl : S) end

12.13 Extension nodes

(Introduced in OCaml 4.02, infix notations for constructs other than expressions added in 4.03, infix notation (e1 ;%ext e2) added in 4.04. )

Extension nodes are generic placeholders in the syntax tree. They are rejected by the type-checker and are intended to be “expanded” by external tools such as -ppx rewriters.

Extension nodes share the same notion of identifier and payload as attributes 12.12.

The first form of extension node is used for “algebraic” categories:

extension::= [% attr-id attr-payload ]
 
expr::= ...
 âˆŁâ€„extension
 
typexpr::= ...
 âˆŁâ€„extension
 
pattern::= ...
 âˆŁâ€„extension
 
module-expr::= ...
 âˆŁâ€„extension
 
module-type::= ...
 âˆŁâ€„extension
 
class-expr::= ...
 âˆŁâ€„extension
 
class-type::= ...
 âˆŁâ€„extension
 

A second form of extension node can be used in structures and signatures, both in the module and object languages:

item-extension::= [%% attr-id attr-payload ]
 
definition::= ...
 âˆŁâ€„item-extension
 
specification::= ...
 âˆŁâ€„item-extension
 
class-field-spec::= ...
 âˆŁâ€„item-extension
 
class-field::= ...
 âˆŁâ€„item-extension
 

An infix form is available for extension nodes when the payload is of the same kind (expression with expression, pattern with pattern ...).

Examples:

let%foo x = 2 in x + 1     === [%foo let x = 2 in x + 1]
begin%foo ... end          === [%foo begin ... end]
x ;%foo 2                  === [%foo x; 2]
module%foo M = ..          === [%%foo module M = ... ]
val%foo x : t              === [%%foo: val x : t]

When this form is used together with the infix syntax for attributes, the attributes are considered to apply to the payload:

fun%foo[@bar] x -> x + 1 === [%foo (fun x -> x + 1)[@bar ] ];

An additional shorthand let%foo x in ... is available for convenience when extension nodes are used to implement binding operators (See 12.23.4).

Furthermore, quoted strings {|...|} can be combined with extension nodes to embed foreign syntax fragments. Those fragments can be interpreted by a preprocessor and turned into OCaml code without requiring escaping quotes. A syntax shortcut is available for them:

{%%foo|...|}               === [%%foo{|...|}]
let x = {%foo|...|}        === let x = [%foo{|...|}]
let y = {%foo bar|...|bar} === let y = [%foo{bar|...|bar}]

For instance, you can use {%sql|...|} to represent arbitrary SQL statements – assuming you have a ppx-rewriter that recognizes the %sql extension.

Note that the word-delimited form, for example {sql|...|sql}, should not be used for signaling that an extension is in use. Indeed, the user cannot see from the code whether this string literal has different semantics than they expect. Moreover, giving semantics to a specific delimiter limits the freedom to change the delimiter to avoid escaping issues.

12.13.1 Built-in extension nodes

(Introduced in OCaml 4.03)

Some extension nodes are understood by the compiler itself:

type t = .. type t += X of int | Y of string let x = [%extension_constructor X] let y = [%extension_constructor Y]
# x <> y;;
- : bool = true

12.14 Extensible variant types

(Introduced in OCaml 4.02)

type-representation::= ...
 âˆŁâ€„= ..
 
specification::= ...
 âˆŁâ€„type [type-params] typeconstr type-extension-spec
 
definition::= ...
 âˆŁâ€„type [type-params] typeconstr type-extension-def
 
type-extension-spec::= += [private] [|] constr-decl { | constr-decl }
 
type-extension-def::= += [private] [|] constr-def { | constr-def }
 
constr-def::= constr-decl
 âˆŁâ€„constr-name = constr
 

Extensible variant types are variant types which can be extended with new variant constructors. Extensible variant types are defined using ... New variant constructors are added using +=.

module Expr = struct type attr = .. type attr += Str of string type attr += | Int of int | Float of float end

Pattern matching on an extensible variant type requires a default case to handle unknown variant constructors:

let to_string = function | Expr.Str s -> s | Expr.Int i -> Int.to_string i | Expr.Float f -> string_of_float f | _ -> "?"

A preexisting example of an extensible variant type is the built-in exn type used for exceptions. Indeed, exception constructors can be declared using the type extension syntax:

type exn += Exc of int

Extensible variant constructors can be rebound to a different name. This allows exporting variants from another module.

# let not_in_scope = Str "Foo";;
Error: Unbound constructor Str
type Expr.attr += Str = Expr.Str
# let now_works = Str "foo";;
val now_works : Expr.attr = Expr.Str "foo"

Extensible variant constructors can be declared private. As with regular variants, this prevents them from being constructed directly by constructor application while still allowing them to be de-structured in pattern-matching.

module B : sig type Expr.attr += private Bool of int val bool : bool -> Expr.attr end = struct type Expr.attr += Bool of int let bool p = if p then Bool 1 else Bool 0 end
# let inspection_works = function | B.Bool p -> (p = 1) | _ -> true;;
val inspection_works : Expr.attr -> bool = <fun>
# let construction_is_forbidden = B.Bool 1;;
Error: Cannot use private constructor Bool to create values of type Expr.attr

12.14.1 Private extensible variant types

(Introduced in OCaml 4.06)

type-representation::= ...
 âˆŁâ€„= private ..
 

Extensible variant types can be declared private. This prevents new constructors from being declared directly, but allows extension constructors to be referred to in interfaces.

module Msg : sig type t = private .. module MkConstr (X : sig type t end) : sig type t += C of X.t end end = struct type t = .. module MkConstr (X : sig type t end) = struct type t += C of X.t end end

12.15 Generative functors

(Introduced in OCaml 4.02)

module-expr::= ...
 âˆŁâ€„functor () -> module-expr
 âˆŁâ€„module-expr ()
 
definition::= ...
 âˆŁâ€„module module-name { ( module-name : module-type ) ∣ () } [ : module-type ]  = module-expr
 
module-type::= ...
 âˆŁâ€„[functor] () -> module-type
 
specification::= ...
 âˆŁâ€„module module-name { ( module-name : module-type ) ∣ () } : module-type
 

A generative functor takes a unit () argument. In order to use it, one must necessarily apply it to this unit argument, ensuring that all type components in the result of the functor behave in a generative way, i.e. they are different from types obtained by other applications of the same functor. This is equivalent to taking an argument of signature sig end, and always applying to struct end, but not to some defined module (in the latter case, applying twice to the same module would return identical types).

As a side-effect of this generativity, one is allowed to unpack first-class modules in the body of generative functors.

12.16 Extension-only syntax

(Introduced in OCaml 4.02.2, extended in 4.03)

Some syntactic constructions are accepted during parsing and rejected during type checking. These syntactic constructions can therefore not be used directly in vanilla OCaml. However, -ppx rewriters and other external tools can exploit this parser leniency to extend the language with these new syntactic constructions by rewriting them to vanilla constructions.

12.16.1 Extension operators

(Introduced in OCaml 4.02.2, extended to unary operators in OCaml 4.12.0)

infix-symbol::= ...
 âˆŁâ€„# { operator-char } # { operator-char ∣ # }
 
prefix-symbol::= ...
 âˆŁâ€„(? ∣ ~ ∣ !) { operator-char } # { operator-char ∣ # }
 

There are two classes of operators available for extensions: infix operators with a name starting with a # character and containing more than one # character, and unary operators with a name (starting with a ?, ~, or ! character) containing at least one # character.

For instance:

# let infix x y = x##y;;
Error: '##' is not a valid value identifier.
# let prefix x = !#x;;
Error: '!#' is not a valid value identifier.

Note that both ## and !# must be eliminated by a ppx rewriter to make this example valid.

12.16.2 Extension literals

(Introduced in OCaml 4.03)

float-literal::= ...
 âˆŁâ€„[-] (0
9) { 0
9 ∣ _ } [. { 0
9 ∣ _ }] [(e ∣ E) [+ ∣ -] (0
9) { 0
9 ∣ _ }] [g
z ∣ G
Z]
 âˆŁâ€„[-] (0x ∣ 0X) (0
9 ∣ A
F ∣ a
f) { 0
9 ∣ A
F ∣ a
f ∣ _ } [. { 0
9 ∣ A
F ∣ a
f ∣ _ }] [(p ∣ P) [+ ∣ -] (0
9) { 0
9 ∣ _ }] [g
z ∣ G
Z]
 
int-literal::= ...
 âˆŁâ€„[-] (0
9) { 0
9 ∣ _ }[g
z ∣ G
Z]
 âˆŁâ€„[-] (0x ∣ 0X) (0
9 ∣ A
F ∣ a
f) { 0
9 ∣ A
F ∣ a
f ∣ _ } [g
z ∣ G
Z]
 âˆŁâ€„[-] (0o ∣ 0O) (0
7) { 0
7 ∣ _ } [g
z ∣ G
Z]
 âˆŁâ€„[-] (0b ∣ 0B) (0
1) { 0
1 ∣ _ } [g
z ∣ G
Z]
 

Int and float literals followed by an one-letter identifier in the range [g..z∣G..Z] are extension-only literals.

12.17 Inline records

(Introduced in OCaml 4.03)

  constr-args::= ...
 âˆŁâ€„record-decl
 

The arguments of sum-type constructors can now be defined using the same syntax as records. Mutable and polymorphic fields are allowed. GADT syntax is supported. Attributes can be specified on individual fields.

Syntactically, building or matching constructors with such an inline record argument is similar to working with a unary constructor whose unique argument is a declared record type. A pattern can bind the inline record as a pseudo-value, but the record cannot escape the scope of the binding and can only be used with the dot-notation to extract or modify fields or to build new constructor values.

type t = | Point of {width: int; mutable x: float; mutable y: float} | Other let v = Point {width = 10; x = 0.; y = 0.} let scale l = function | Point p -> Point {p with x = l *. p.x; y = l *. p.y} | Other -> Other let print = function | Point {x; y; _} -> Printf.printf "%f/%f" x y | Other -> () let reset = function | Point p -> p.x <- 0.; p.y <- 0. | Other -> ()
let invalid = function | Point p -> p
Error: This form is not allowed as the type of the inlined record could escape.

12.18 Documentation comments

(Introduced in OCaml 4.03)

Comments which start with ** are treated specially by the compiler. They are automatically converted during parsing into attributes (see 12.12) to allow tools to process them as documentation.

Such comments can take three forms: floating comments, item comments and label comments. Any comment starting with ** which does not match one of these forms will cause the compiler to emit warning 50.

Comments which start with ** are also used by the ocamldoc documentation generator (see 19). The three comment forms recognised by the compiler are a subset of the forms accepted by ocamldoc (see 19.2).

12.18.1 Floating comments

Comments surrounded by blank lines that appear within structures, signatures, classes or class types are converted into floating-attributes. For example:

type t = T (** Now some definitions for [t] *) let mkT = T

will be converted to:

type t = T [@@@ocaml.text " Now some definitions for [t] "] let mkT = T

12.18.2 Item comments

Comments which appear immediately before or immediately after a structure item, signature item, class item or class type item are converted into item-attributes. Immediately before or immediately after means that there must be no blank lines, ;;, or other documentation comments between them. For example:

type t = T (** A description of [t] *)

or

(** A description of [t] *) type t = T

will be converted to:

type t = T [@@ocaml.doc " A description of [t] "]

Note that, if a comment appears immediately next to multiple items, as in:

type t = T (** An ambiguous comment *) type s = S

then it will be attached to both items:

type t = T [@@ocaml.doc " An ambiguous comment "] type s = S [@@ocaml.doc " An ambiguous comment "]

and the compiler will emit warning 50.

12.18.3 Label comments

Comments which appear immediately after a labelled argument, record field, variant constructor, object method or polymorphic variant constructor are are converted into attributes. Immediately after means that there must be no blank lines or other documentation comments between them. For example:

type t1 = lbl:int (** Labelled argument *) -> unit type t2 = { fld: int; (** Record field *) fld2: float; } type t3 = | Cstr of string (** Variant constructor *) | Cstr2 of string type t4 = < meth: int * int; (** Object method *) > type t5 = [ `PCstr (** Polymorphic variant constructor *) ]

will be converted to:

type t1 = lbl:(int [@ocaml.doc " Labelled argument "]) -> unit type t2 = { fld: int [@ocaml.doc " Record field "]; fld2: float; } type t3 = | Cstr of string [@ocaml.doc " Variant constructor "] | Cstr2 of string type t4 = < meth : int * int [@ocaml.doc " Object method "] > type t5 = [ `PCstr [@ocaml.doc " Polymorphic variant constructor "] ]

Note that label comments take precedence over item comments, so:

type t = T of string (** Attaches to T not t *)

will be converted to:

type t = T of string [@ocaml.doc " Attaches to T not t "]

whilst:

type t = T of string (** Attaches to T not t *) (** Attaches to t *)

will be converted to:

type t = T of string [@ocaml.doc " Attaches to T not t "] [@@ocaml.doc " Attaches to t "]

In the absence of meaningful comment on the last constructor of a type, an empty comment (**) can be used instead:

type t = T of string (**) (** Attaches to t *)

will be converted directly to

type t = T of string [@@ocaml.doc " Attaches to t "]

12.19 Extended indexing operators

(Introduced in 4.06)

dot-ext::= 
 âˆŁâ€„dot-operator-char { operator-char }
 
dot-operator-char::= ! ∣ ? ∣ core-operator-char ∣ % ∣ :
 
expr::= ...
 âˆŁâ€„expr . [module-path .] dot-ext ( ( expr ) ∣ [ expr ] ∣ { expr } ) [ <- expr ]
 
operator-name::= ...
 âˆŁâ€„. dot-ext (() ∣ [] ∣ {}) [<-]
 

This extension provides syntactic sugar for getting and setting elements for user-defined indexed types. For instance, we can define python-like dictionaries with

module Dict = struct include Hashtbl let ( .%{} ) tabl index = find tabl index let ( .%{}<- ) tabl index value = add tabl index value end let dict = let dict = Dict.create 10 in let () = dict.Dict.%{"one"} <- 1; let open Dict in dict.%{"two"} <- 2 in dict
# dict.Dict.%{"one"};;
- : int = 1
# let open Dict in dict.%{"two"};;
- : int = 2

12.19.1 Multi-index notation

expr::= ...
 âˆŁâ€„expr . [module-path .] dot-ext ( expr { ; expr }+ ) [ <- expr ]
 âˆŁâ€„expr . [module-path .] dot-ext [ expr { ; expr }+ ] [ <- expr ]
 âˆŁâ€„expr . [module-path .] dot-ext { expr { ; expr }+ } [ <- expr ]
 
operator-name::= ...
 âˆŁâ€„. dot-ext ((;..) ∣ [;..] ∣ {;..}) [<-]
 

Multi-index are also supported through a second variant of indexing operators

let (.%[;..]) = Bigarray.Genarray.get let (.%{;..}) = Bigarray.Genarray.get let (.%(;..)) = Bigarray.Genarray.get

which is called when an index literals contain a semicolon separated list of expressions with two and more elements:

let sum x y = x.%[1;2;3] + y.%[1;2] (* is equivalent to *) let sum x y = (.%[;..]) x [|1;2;3|] + (.%[;..]) y [|1;2|]

In particular this multi-index notation makes it possible to uniformly handle indexing Genarray and other implementations of multidimensional arrays.

module A = Bigarray.Genarray let (.%{;..}) = A.get let (.%{;..}<- ) = A.set let (.%{ }) a k = A.get a [|k|] let (.%{ }<-) a k x = A.set a [|k|] x let syntax_compare vec mat t3 t4 = vec.%{0} = A.get vec [|0|] && mat.%{0;0} = A.get mat [|0;0|] && t3.%{0;0;0} = A.get t3 [|0;0;0|] && t4.%{0;0;0;0} = t4.{0,0,0,0}

Beware that the differentiation between the multi-index and single index operators is purely syntactic: multi-index operators are restricted to index expressions that contain one or more semicolons ;. For instance,

let pair vec mat = vec.%{0}, mat.%{0;0}

is equivalent to

let pair vec mat = (.%{ }) vec 0, (.%{;..}) mat [|0;0|]

Notice that in the vec case, we are calling the single index operator, (.%{}), and not the multi-index variant, (.{;..}). For this reason, it is expected that most users of multi-index operators will need to define conjointly a single index variant

let (.%{;..}) = A.get let (.%{ }) a k = A.get a [|k|]

to handle both cases uniformly.

12.20 Empty variant types

(Introduced in 4.07.0)

type-representation::= ...
 âˆŁâ€„= |

This extension allows user to define empty variants. Empty variant type can be eliminated by refutation case of pattern matching.

type t = | let f (x: t) = match x with _ -> .

12.21 Alerts

(Introduced in 4.08)

Since OCaml 4.08, it is possible to mark components (such as value or type declarations) in signatures with “alerts” that will be reported when those components are referenced. This generalizes the notion of “deprecated” components which were previously reported as warning 3. Those alerts can be used for instance to report usage of unsafe features, or of features which are only available on some platforms, etc.

Alert categories are identified by a symbolic identifier (a lowercase identifier, following the usual lexical rules) and an optional message. The identifier is used to control which alerts are enabled, and which ones are turned into fatal errors. The message is reported to the user when the alert is triggered (i.e. when the marked component is referenced).

The ocaml.alert or alert attribute serves two purposes: (i) to mark component with an alert to be triggered when the component is referenced, and (ii) to control which alert names are enabled. In the first form, the attribute takes an identifier possibly followed by a message. Here is an example of a value declaration marked with an alert:

module U: sig
  val fork: unit -> bool
    [@@alert unix "This function is only available under Unix."]
end

Here unix is the identifier for the alert. If this alert category is enabled, any reference to U.fork will produce a message at compile time, which can be turned or not into a fatal error.

And here is another example as a floating attribute on top of an “.mli” file (i.e. before any other non-attribute item) or on top of an “.ml” file without a corresponding interface file, so that any reference to that unit will trigger the alert:

[@@@alert unsafe "This module is unsafe!"]

Controlling which alerts are enabled and whether they are turned into fatal errors is done either through the compiler’s command-line option -alert <spec> or locally in the code through the alert or ocaml.alert attribute taking a single string payload <spec>. In both cases, the syntax for <spec> is a concatenation of items of the form:

As a special case, if id is all, it stands for all alerts.

Here are some examples:

(* Disable all alerts, reenables just unix (as a soft alert) and window
   (as a fatal-error), for the rest of the current structure *)

[@@@alert "-all--all+unix@window"]
 ...

let x =
  (* Locally disable the window alert *)
  begin[@alert "-window"]
      ...
  end

Before OCaml 4.08, there was support for a single kind of deprecation alert. It is now known as the deprecated alert, but legacy attributes to trigger it and the legacy ways to control it as warning 3 are still supported. For instance, passing -w +3 on the command-line is equivalent to -alert +deprecated, and:

val x: int
  [@@ocaml.deprecated "Please do something else"]

is equivalent to:

val x: int
  [@@ocaml.alert deprecated "Please do something else"]

12.22 Generalized open statements

(Introduced in 4.08)

definition::= ...
 âˆŁâ€„ open  module-expr
 âˆŁâ€„ open! module-expr
 
specification::= ...
 âˆŁâ€„ open  extended-module-path
 âˆŁâ€„ open! extended-module-path
 
expr::= ...
 âˆŁâ€„let open  module-expr in expr
 âˆŁâ€„let open! module-expr in expr
 

This extension makes it possible to open any module expression in module structures and expressions. A similar mechanism is also available inside module types, but only for extended module paths (e.g. F(X).G(Y)).

For instance, a module can be constrained when opened with

module M = struct let x = 0 let hidden = 1 end open (M:sig val x: int end) let y = hidden
Error: Unbound value hidden

Another possibility is to immediately open the result of a functor application

let sort (type x) (x:x list) = let open Set.Make(struct type t = x let compare=compare end) in elements (of_list x)
val sort : 'x list -> 'x list = <fun>

Going further, this construction can introduce local components inside a structure,

module M = struct let x = 0 open! struct let x = 0 let y = 1 end let w = x + y end
module M : sig val x : int val w : int end

One important restriction is that types introduced by open struct ... end cannot appear in the signature of the enclosing structure, unless they are defined equal to some non-local type. So:

module M = struct open struct type 'a t = 'a option = None | Some of 'a end let x : int t = Some 1 end
module M : sig val x : int option end

is OK, but:

module M = struct open struct type t = A end let x = A end
Error: The type t introduced by this open appears in the signature. The value x has no valid type if t is hidden.

is not because x cannot be given any type other than t, which only exists locally. Although the above would be OK if x too was local:

module M: sig end = struct open struct type t = A end 
 open struct let x = A end 
 end
module M : sig end

Inside signatures, extended opens are limited to extended module paths,

module type S = sig module F: sig end -> sig type t end module X: sig end open F(X) val f: t end
module type S = sig module F : sig end -> sig type t end module X : sig end val f : F(X).t end

and not

  open struct type t = int end

In those situations, local substitutions(see 12.7.2) can be used instead.

Beware that this extension is not available inside class definitions:

class c =
  let open Set.Make(Int) in
  ...

12.23 Binding operators

(Introduced in 4.08.0)

let-operator::= 
 âˆŁâ€„let (core-operator-char ∣ <) { dot-operator-char }
 
and-operator::= 
 âˆŁâ€„and (core-operator-char ∣ <) { dot-operator-char }
 
operator-name ::= ...
 âˆŁâ€„let-operator
 âˆŁâ€„and-operator
 
letop-binding ::= pattern = expr
 âˆŁâ€„value-name
 
expr::= ...
 âˆŁâ€„let-operator letop-binding { and-operator letop-binding } in expr
 

Binding operators offer syntactic sugar to expose library functions under (a variant of) the familiar syntax of standard keywords. Currently supported “binding operators” are let<op> and and<op>, where <op> is an operator symbol, for example and+$.

Binding operators were introduced to offer convenient syntax for working with monads and applicative functors; for those, we propose conventions using operators * and + respectively. They may be used for other purposes, but one should keep in mind that each new unfamiliar notation introduced makes programs harder to understand for non-experts. We expect that new conventions will be developed over time on other families of operator.

12.23.1 Examples

Users can define let operators:

let ( let* ) o f = match o with | None -> None | Some x -> f x let return x = Some x
val ( let* ) : 'a option -> ('a -> 'b option) -> 'b option = <fun> val return : 'a -> 'a option = <fun>

and then apply them using this convenient syntax:

let find_and_sum tbl k1 k2 = let* x1 = Hashtbl.find_opt tbl k1 in let* x2 = Hashtbl.find_opt tbl k2 in return (x1 + x2)
val find_and_sum : ('a, int) Hashtbl.t -> 'a -> 'a -> int option = <fun>

which is equivalent to this expanded form:

let find_and_sum tbl k1 k2 = ( let* ) (Hashtbl.find_opt tbl k1) (fun x1 -> ( let* ) (Hashtbl.find_opt tbl k2) (fun x2 -> return (x1 + x2)))
val find_and_sum : ('a, int) Hashtbl.t -> 'a -> 'a -> int option = <fun>

Users can also define and operators:

module ZipSeq = struct type 'a t = 'a Seq.t open Seq let rec return x = fun () -> Cons(x, return x) let rec prod a b = fun () -> match a (), b () with | Nil, _ | _, Nil -> Nil | Cons(x, a), Cons(y, b) -> Cons((x, y), prod a b) let ( let+ ) f s = map s f let ( and+ ) a b = prod a b end
module ZipSeq : sig type 'a t = 'a Seq.t val return : 'a -> 'a Seq.t val prod : 'a Seq.t -> 'b Seq.t -> ('a * 'b) Seq.t val ( let+ ) : 'a Seq.t -> ('a -> 'b) -> 'b Seq.t val ( and+ ) : 'a Seq.t -> 'b Seq.t -> ('a * 'b) Seq.t end

to support the syntax:

open ZipSeq let sum3 z1 z2 z3 = let+ x1 = z1 and+ x2 = z2 and+ x3 = z3 in x1 + x2 + x3
val sum3 : int Seq.t -> int Seq.t -> int Seq.t -> int Seq.t = <fun>

which is equivalent to this expanded form:

open ZipSeq let sum3 z1 z2 z3 = ( let+ ) (( and+ ) (( and+ ) z1 z2) z3) (fun ((x1, x2), x3) -> x1 + x2 + x3)
val sum3 : int Seq.t -> int Seq.t -> int Seq.t -> int Seq.t = <fun>

12.23.2 Conventions

An applicative functor should provide a module implementing the following interface:

module type Applicative_syntax = sig type 'a t val ( let+ ) : 'a t -> ('a -> 'b) -> 'b t val ( and+ ): 'a t -> 'b t -> ('a * 'b) t end

where (let+) is bound to the map operation and (and+) is bound to the monoidal product operation.

A monad should provide a module implementing the following interface:

module type Monad_syntax = sig include Applicative_syntax val ( let* ) : 'a t -> ('a -> 'b t) -> 'b t val ( and* ): 'a t -> 'b t -> ('a * 'b) t end

where (let*) is bound to the bind operation, and (and*) is also bound to the monoidal product operation.

12.23.3 General desugaring rules

The form

let<op0>
  x1 = e1
and<op1>
  x2 = e2
and<op2>
  x3 = e3
in e

desugars into

( let<op0> )
  (( and<op2> )
    (( and<op1> )
      e1
      e2)
    e3)
  (fun ((x1, x2), x3) -> e)

This of course works for any number of nested and-operators. One can express the general rule by repeating the following simplification steps:

Note that the grammar allows mixing different operator symbols in the same binding (<op0>, <op1>, <op2> may be distinct), but we strongly recommend APIs where let-operators and and-operators working together use the same symbol.

12.23.4 Short notation for variable bindings (let-punning)

(Introduced in 4.13.0)

When the expression being bound is a variable, it can be convenient to use the shorthand notation let+ x in ..., which expands to let+ x = x in .... This notation, also known as let-punning, allows the sum3 function above can be written more concisely as:

open ZipSeq let sum3 z1 z2 z3 = let+ z1 and+ z2 and+ z3 in z1 + z2 + z3
val sum3 : int Seq.t -> int Seq.t -> int Seq.t -> int Seq.t = <fun>

This notation is also supported for extension nodes, expanding let%foo x in ... to let%foo x = x in .... However, to avoid confusion, this notation is not supported for plain let bindings.

12.24 Effect handlers

(Introduced in 5.0)

Note: Effect handlers in OCaml 5.0 should be considered experimental. Effect handlers are exposed in the standard library’s Effect module as a thin wrapper around their implementation in the runtime. They are not supported as a language feature with new syntax. You can rely on them to build non-local control-flow abstractions such as user-level threading that do not expose the effect handler primitives to the user. Expect breaking changes in the future.

Effect handlers are a mechanism for modular programming with user-defined effects. Effect handlers allow the programmers to describe computations that perform effectful operations, whose meaning is described by handlers that enclose the computations. Effect handlers are a generalization of exception handlers and enable non-local control-flow mechanisms such as resumable exceptions, lightweight threads, coroutines, generators and asynchronous I/O to be composably expressed. In this tutorial, we shall see how some of these mechanisms can be built using effect handlers.

12.24.1 Basics

To understand the basics, let us define an effect (that is, an operation) that takes an integer argument and returns an integer result. We name this effect Xchg.

open Effect open Effect.Deep type _ Effect.t += Xchg: int -> int t let comp1 () = perform (Xchg 0) + perform (Xchg 1)

We declare the exchange effect Xchg by extending the pre-defined extensible variant type Effect.t with a new constructor Xchg: int -> int t. The declaration may be intuitively read as “the Xchg effect takes an integer parameter, and when this effect is performed, it returns an integer”. The computation comp1 performs the effect twice using the perform primitive and returns their sum.

We can handle the Xchg effect by implementing a handler that always returns the successor of the offered value:

try_with comp1 () { effc = fun (type a) (eff: a t) -> match eff with | Xchg n -> Some (fun (k: (a, _) continuation) -> continue k (n+1)) | _ -> None }
- : int = 3

try_with runs the computation comp1 () under an effect handler that handles the Xchg effect. As mentioned earlier, effect handlers are a generalization of exception handlers. Similar to exception handlers, when the computation performs the Xchg effect, the control jumps to the corresponding handler. However, unlike exception handlers, the handler is also provided with the delimited continuation k, which represents the suspended computation between the point of perform and this handler.

The handler uses the continue primitive to resume the suspended computation with the successor of the offered value. In this example, the computation comp1 performs Xchg 0 and Xchg 1 and receives the values 1 and 2 from the handler respectively. Hence, the whole expression evaluates to 3.

It is useful to note that we must use a locally abstract type (type a) in the effect handler. The type Effect.t is a GADT, and the effect declarations may have different type parameters for different effects. The type parameter a in the type a Effect.t represents the type of the value returned when performing the effect. From the fact that eff has type a Effect.t and from the fact that Xchg n has type int Effect.t, the type-checker deduces that a must be int, which is why we are allowed to pass the integer value n+1 as an argument to continue k.

Another point to note is that the catch-all case “| _ -> None” is necessary when handling effects. This case may be intuitively read as “forward the unhandled effects to the outer handler”.

In this example, we use the deep version of the effect handlers here as opposed to the shallow version. A deep handler monitors a computation until the computation terminates (either normally or via an exception), and handles all of the effects performed (in sequence) by the computation. In contrast, a shallow handler monitors a computation until either the computation terminates or the computation performs one effect, and it handles this single effect only. In situations where they are applicable, deep handlers are usually preferred. An example that utilises shallow handlers is discussed later in 12.24.6.

12.24.2 Concurrency

The expressive power of effect handlers comes from the delimited continuation. While the previous example immediately resumed the computation, the computation may be resumed later, running some other computation in the interim. Let us extend the previous example and implement message-passing concurrency between two concurrent computations using the Xchg effect. We call these concurrent computations tasks.

A task either is in a suspended state or is completed. We represent the task status as follows:

type 'a status = Complete of 'a | Suspended of {msg: int; cont: (int, 'a status) continuation}

A task either is complete, with a result of type 'a, or is suspended with the message msg to send and the continuation cont. The type (int,'a status) continuation says that the suspended computation expects an int value to resume and returns a 'a status value when resumed.

Next, we define a step function that executes one step of computation until it completes or suspends:

let step (f : unit -> 'a) () : 'a status = match_with f () { retc = (fun v -> Complete v); exnc = raise; effc = fun (type a) (eff: a t) -> match eff with | Xchg msg -> Some (fun (cont: (a, _) continuation) -> Suspended {msg; cont}) | _ -> None }

The argument to the step function, f, is a computation that can perform an Xchg effect and returns a result of type 'a. The step function itself returns a 'a status value.

In the step function, we use the match_with primitive. Like try_with, match_with primitive installs an effect handler. However, unlike try_with, where only the effect case effc is provided, match_with expects the handlers for the value (retc) and exceptional (exnc) return cases. In fact, try_with can be defined using match_with as follows: let try_with f v {effc} = match_with f v {retc = Fun.id; exnc = raise; effc}.

In the step function,

Since the step function handles the Xchg effect, step f is a computation that does not perform the Xchg effect. It may however perform other effects. Moreover, since we are using deep handlers, the continuation cont stored in the status does not perform the Xchg effect.

We can now write a simple scheduler that runs a pair of tasks to completion:

let rec run_both a b = match a (), b () with | Complete va, Complete vb -> (va, vb) | Suspended {msg = m1; cont = k1}, Suspended {msg = m2; cont = k2} -> run_both (fun () -> continue k1 m2) (fun () -> continue k2 m1) | _ -> failwith "Improper synchronization"

Both of the tasks may run to completion, or both may offer to exchange a message. In the latter case, each computation receives the value offered by the other computation. The situation where one computation offers an exchange while the other computation terminates is regarded as a programmer error, and causes the handler to raise an exception

We can now define a second computation that also exchanges two messages:

let comp2 () = perform (Xchg 21) * perform (Xchg 21)

Finally, we can run the two computations together:

run_both (step comp1) (step comp2)
- : int * int = (42, 0)

The computation comp1 offers the values 0 and 1 and in exchange receives the values 21 and 21, which it adds, producing 42. The computation comp2 offers the values 21 and 21 and in exchange receives the values 0 and 1, which it multiplies, producing 0. The communication between the two computations is programmed entirely inside run_both. Indeed, the definitions of comp1 and comp2, alone, do not assign any meaning to the Xchg effect.

12.24.3 User-level threads

Let us extend the previous example for an arbitrary number of tasks. Many languages such as GHC Haskell and Go provide user-level threads as a primitive feature implemented in the runtime system. With effect handlers, user-level threads and their schedulers can be implemented in OCaml itself. Typically, user-level threading systems provide a fork primitive to spawn off a new concurrent task and a yield primitive to yield control to some other task. Correspondingly, we shall declare two effects as follows:

type _ Effect.t += Fork : (unit -> unit) -> unit t | Yield : unit t

The Fork effect takes a thunk (a suspended computation, represented as a function of type unit -> unit) and returns a unit to the performer. The Yield effect is unparameterized and returns a unit when performed. Let us consider that a task performing an Xchg effect may match with any other task also offering to exchange a value.

We shall also define helper functions that simply perform these effects:

let fork f = perform (Fork f) let yield () = perform Yield let xchg v = perform (Xchg v)

A top-level run function defines the scheduler:

(* A concurrent round-robin scheduler *) let run (main : unit -> unit) : unit = let exchanger = ref None in (* waiting exchanger *) let run_q = Queue.create () in (* scheduler queue *) let enqueue k v = let task () = continue k v in Queue.push task run_q in let dequeue () = if Queue.is_empty run_q then () (* done *) else begin let task = Queue.pop run_q in task () end in let rec spawn (f : unit -> unit) : unit = match_with f () { retc = dequeue; exnc = (fun e -> print_endline (Printexc.to_string e); dequeue ()); effc = fun (type a) (eff : a t) -> match eff with | Yield -> Some (fun (k : (a, unit) continuation) -> enqueue k (); dequeue ()) | Fork f -> Some (fun (k : (a, unit) continuation) -> enqueue k (); spawn f) | Xchg n -> Some (fun (k : (int, unit) continuation) -> begin match !exchanger with | Some (n', k') -> exchanger := None; enqueue k' n; continue k n' | None -> exchanger := Some (n, k); dequeue () end) | _ -> None } in spawn main

We use a mutable queue run_q to hold the scheduler queue. The FIFO queue enables round-robin scheduling of tasks in the scheduler. enqueue inserts tasks into the queue, and dequeue extracts tasks from the queue and runs them. The reference cell exchanger holds a (suspended) task offering to exchange a value. At any time, there is either zero or one suspended task that is offering an exchange.

The heavy lifting is done by the spawn function. The spawn function runs the given computation f in an effect handler. If f returns with a value (case retc), we dequeue and run the next task from the scheduler queue. If the computation f raises an exception (case exnc), we print the exception and run the next task from the scheduler.

The computation f may also perform effects. If f performs the Yield effect, the current task is suspended (inserted into the queue of ready tasks), and the next task from the scheduler queue is run. If the effect is Fork f, then the current task is suspended, and the new task f is executed immediately via a tail call to spawn f. Note that this choice to run the new task first is arbitrary. We could very well have chosen instead to insert the task for f into the ready queue and resumed k immediately.

If the effect is Xchg, then we first check whether there is a task waiting to exchange. If so, we enqueue the waiting task with the current value being offered and immediately resume the current task with the value being offered. If not, we make the current task the waiting exchanger, and run the next task from the scheduler queue.

Note that this scheduler code is not perfect – it can leak resources. We shall explain and fix this in the next section 12.24.3.

Now we can write a concurrent program that utilises the newly defined operations:

open Printf let _ = run (fun _ -> fork (fun _ -> printf "[t1] Sending 0\n"; let v = xchg 0 in printf "[t1] received %d\n" v); fork (fun _ -> printf "[t2] Sending 1\n"; let v = xchg 1 in printf "[t2] received %d\n" v))
[t1] Sending 0 [t2] Sending 1 [t2] received 0 [t1] received 1

Observe that the messages from the two tasks are interleaved. Notice also that the snippet above makes no reference to the effect handlers and is in direct style (no monadic operations). This example illustrates that, with effect handlers, the user code in a concurrent program can remain in simple direct style, and the use of effect handlers can be fully contained within the concurrency library implementation.

Resuming with an exception

In addition to resuming a continuation with a value, effect handlers also permit resuming by raising an effect at the point of perform. This is done with the help of the discontinue primitive. The discontinue primitive helps ensure that resources are always eventually deallocated, even in the presence of effects.

For example, consider the dequeue operation in the previous example reproduced below:


 let dequeue () = if Queue.is_empty run_q then () (* done *) else (Queue.pop run_q) ()

If the scheduler queue is empty, dequeue considers that the scheduler is done and returns to the caller. However, there may still be a task waiting to exchange a value (stored in the reference cell exchanger), which remains blocked forever! If the blocked task holds onto resources, these resources are leaked. For example, consider the following task:

let leaky_task () = fork (fun _ -> let oc = open_out "secret.txt" in Fun.protect ~finally:(fun _ -> close_out oc) (fun _ -> output_value oc (xchg 0)))

The task writes the received message to the file secret.txt. It uses Fun.protect to ensure that the output channel oc is closed on both normal and exceptional return cases. Unfortunately, this is not sufficient. If the exchange effect xchg 0 cannot be matched with an exchange effect performed by some other thread, then this task remains blocked forever. Thus, the output channel oc is never closed.

To avoid this problem, one must adhere to a simple discipline: every continuation must be eventually either continued or discontinued. Here, we use discontinue to ensure that the blocked task does not remain blocked forever. By discontinuing this task, we force it to terminate (with an exception):

exception Improper_synchronization let dequeue () = if Queue.is_empty run_q then begin match !exchanger with | None -> () (* done *) | Some (n, k) -> exchanger := None; discontinue k Improper_synchronization end else (Queue.pop run_q) ()

When the scheduler queue is empty and there is a blocked exchanger thread, the dequeue function discontinues the blocked thread with an Improper_synchronization exception. This exception is raised at the blocked xchg function call, which causes the finally block to be run and closes the output channel oc. From the point of view of the user, it seems as though the function call xchg 0 raises the exception Improper_synchronization.

12.24.4 Control inversion

When it comes to performing traversals on a data structure, there are two fundamental ways depending on whether the producer or the consumer has the control over the traversal. For example, in List.iter f l, the producer List.iter has the control and pushes the element to the consumer f who processes them. On the other hand, the Seq module provides a mechanism similar to delayed lists where the consumer controls the traversal. For example, Seq.forever Random.bool returns an infinite sequence of random bits where every bit is produced (on demand) when queried by the consumer.

Naturally, producers such as List.iter are easier to write in the former style. The latter style is ergonomically better for the consumer since it is preferable and more natural to be in control. To have the best of both worlds, we would like to write a producer in the former style and automatically convert it to the latter style. The conversion can be written once and for all as a library function, thanks to effect handlers. Let us name this function invert. We will first look at how to use the invert function before looking at its implementation details. The type of this function is given below:

val invert : iter:(('a -> unit) -> unit) -> 'a Seq.t

The invert function takes an iter function (a producer that pushes elements to the consumer) and returns a sequence (where the consumer has the control). For example,

let lst_iter = Fun.flip List.iter [1;2;3]
val lst_iter : (int -> unit) -> unit = <fun>

is an iter function with type (int -> unit) -> unit. The expression lst_iter f pushes the elements 1, 2 and 3 to the consumer f. For example,

lst_iter (fun i -> Printf.printf "%d\n" i)
1 2 3 - : unit = ()

The expression invert lst_iter returns a sequence that allows the consumer to traverse the list on demand. For example,

let s = invert ~iter:lst_iter let next = Seq.to_dispenser s;;
val s : int Seq.t = <fun> val next : unit -> int option = <fun>
next();;
- : int option = Some 1
next();;
- : int option = Some 2
next();;
- : int option = Some 3
next();;
- : int option = None

We can use the same invert function on any iter function. For example,

let s = invert ~iter:(Fun.flip String.iter "OCaml") let next = Seq.to_dispenser s;;
val s : char Seq.t = <fun> val next : unit -> char option = <fun>
next();;
- : char option = Some 'O'
next();;
- : char option = Some 'C'
next();;
- : char option = Some 'a'
next();;
- : char option = Some 'm'
next();;
- : char option = Some 'l'
next();;
- : char option = None

Implementing control inversion

The implementation of the invert function is given below:

let invert (type a) ~(iter : (a -> unit) -> unit) : a Seq.t = let module M = struct type _ Effect.t += Yield : a -> unit t end in let yield v = perform (M.Yield v) in fun () -> match_with iter yield { retc = (fun _ -> Seq.Nil); exnc = raise; effc = fun (type b) (eff : b Effect.t) -> match eff with | M.Yield v -> Some (fun (k: (b,_) continuation) -> Seq.Cons (v, continue k)) | _ -> None }

The invert function declares an effect Yield that takes the element to be yielded as a parameter. The yield function performs the Yield effect. The lambda abstraction fun () -> ... delays all action until the first element of the sequence is demanded. Once this happens, the computation iter yield is executed under an effect handler. Every time the iter function pushes an element to the yield function, the computation is interrupted by the Yield effect. The Yield effect is handled by returning the value Seq.Cons(v,continue k) to the consumer. The consumer gets the element v as well as the suspended computation, which in the consumer’s eyes is just the tail of sequence.

When the consumer demands the next element from the sequence (by applying it to ()), the continuation k is resumed. This allows the computation iter yield to make progress, until it either yields another element or terminates normally. In the latter case, the value Seq.Nil is returned, indicating to the consumer that the iteration is over.

It is important to note that the sequence returned by the invert function is ephemeral (as defined by the Seq module) i.e., the sequence must be used at most once. Additionally, the sequence must be fully consumed (i.e., used at least once) so as to ensure that the captured continuation is used linearly.

12.24.5 Semantics

In this section, we shall see the semantics of effect handlers with the help of examples.

Nesting handlers

Like exception handlers, effect handlers can be nested.

type _ Effect.t += E : int t | F : string t let foo () = perform F let bar () = try_with foo () { effc = fun (type a) (eff: a t) -> match eff with | E -> Some (fun (k: (a,_) continuation) -> failwith "impossible") | _ -> None } let baz () = try_with bar () { effc = fun (type a) (eff: a t) -> match eff with | F -> Some (fun (k: (a,_) continuation) -> continue k "Hello, world!") | _ -> None }

In this example, the computation foo performs F, the inner handler handles only E and the outer handler handles F. The call to baz returns Hello, world!.

baz ()
- : string = "Hello, world!"

Fibers

It is useful to know a little bit about the implementation of effect handlers to appreciate the design choices and their performance characteristics. Effect handlers are implemented with the help of runtime-managed, dynamically growing segments of stack called fibers. The program stack in OCaml is a linked list of such fibers.

A new fiber is allocated for evaluating the computation enclosed by an effect handler. The fiber is freed when the computation returns to the caller either normally by returning a value or by raising an exception.

At the point of perform in foo in the previous example, the program stack looks like this:

+-----+ +-----+ +-----+ | | | | | | | baz |<--| bar |<--| foo | | | | | | | | | | | | | +-----+ +-----+ +-----+ <- stack_pointer

The two links correspond to the two effect handlers in the program. When the effect F is handled in baz, the program state looks as follows:

+-----+ +-----+ +-----+ | | | | | | +-+ | baz | | bar |<--| foo |<--|k| | | | | | | +-+ +-----+ <- stack_pointer +-----+ +-----+

The delimited continuation k is an object on the heap that refers to the segment of the stack that corresponds to the suspended computation. Capturing a continuation does not involve copying stack frames. When the continuation is resumed, the stack is restored to the previous state by linking together the segment pointed to by k to the current stack. Since neither continuation capture nor resumption requires copying stack frames, suspending the execution using perform and resuming it using either continue or discontinue are fast.

Unhandled effects

Unlike languages such as Eff and Koka, effect handlers in OCaml do not provide effect safety; the compiler does not statically ensure that all the effects performed by the program are handled. If effects do not have a matching handler, then an Effect.Unhandled exception is raised at the point of the corresponding perform. For example, in the previous example, bar does not handle the effect F. Hence, we will get an Effect.Unhandled F exception when we run bar.

try bar () with Effect.Unhandled F -> "Saw Effect.Unhandled exception"
- : string = "Saw Effect.Unhandled exception"

Linear continuations

As discussed earlier 12.24.3, the delimited continuations in OCaml must be used linearly – every captured continuation must be resumed either with a continue or discontinue exactly once. Attempting to use a continuation more than once raises a Continuation_already_resumed exception. For example:

try_with perform (Xchg 0) { effc = fun (type a) (eff : a t) -> match eff with | Xchg n -> Some (fun (k: (a, _) continuation) -> continue k 21 + continue k 21) | _ -> None }
Exception: Stdlib.Effect.Continuation_already_resumed.

The primary motivation for adding effect handlers to OCaml is to enable concurrent programming. One-shot continuations are sufficient for almost all concurrent programming needs. They are also much cheaper to implement compared to multi-shot continuations since they do not require stack frames to be copied. Moreover, OCaml programs may also manipulate linear resources such as sockets and file descriptors. The linearity discipline is easily broken if the continuations are allowed to resume more than once. It would be quite hard to debug such linearity violations on resources due to the lack of static checks for linearity and the non-local nature of control flow. Hence, OCaml does not support multi-shot continuations.

While the “at most once resumption” property of continuations is ensured with a dynamic check, there is no check to ensure that the continuations are resumed “at least once”. It is left to the user to ensure that the captured continuations are resumed at least once. Not resuming continuations will leak the memory allocated for the fibers as well as any resources that the suspended computation may hold.

One may install a finaliser on the captured continuation to ensure that the resources are freed:

exception Unwind Gc.finalise (fun k -> try ignore (discontinue k Unwind) with _ -> ()) k

In this case, if k becomes unreachable, then the finaliser ensures that the continuation stack is unwound by discontinuing with an Unwind exception, allowing the computation to free up resources. However, the runtime cost of finalisers is much more than the cost of capturing a continuation. Hence, it is recommended that the user take care of resuming the continuation exactly once rather than relying on the finaliser.

12.24.6 Shallow handlers

The examples that we have seen so far have used deep handlers. A deep handler handles all the effects performed (in sequence) by the computation. Whenever a continuation is captured in a deep handler, the captured continuation also includes the handler. This means that, when the continuation is resumed, the effect handler is automatically re-installed, and will handle the effect(s) that the computation may perform in the future.

OCaml also provides shallow handlers. Compared to deep handlers, a shallow handler handles only the first effect performed by the computation. The continuation captured in a shallow handler does not include the handler. This means that, when the continuation is resumed, the handler is no longer present. For this reason, when the continuation is resumed, the user is expected to provide a new effect handler (possibly a different one) to handle the next effect that the computation may perform.

Shallow handlers make it easier to express certain kinds of programs. Let us implement a shallow handler that enforces a particular sequence of effects (a protocol) on a computation. For this example, let us consider that the computation may perform the following effects:

type _ Effect.t += Send : int -> unit Effect.t | Recv : int Effect.t

Let us assume that we want to enforce a protocol that only permits an alternating sequence of Send and Recv effects that conform to the regular expression (Send;Recv)*;Send?. Hence, the sequence of effects [] (the empty sequence), [Send], [Send;Recv], [Send;Recv;Send], etc., are allowed, but not [Recv], [Send;Send], [Send;Recv;Recv], etc. The key observation here is that the set of effects handled evolves over time. We can enforce this protocol quite naturally using shallow handlers as shown below:

open Effect.Shallow let run (comp: unit -> unit) : unit = let rec loop_send : type a. (a,unit) continuation -> a -> unit = fun k v -> continue_with k v { retc = Fun.id; exnc = raise; effc = fun (type b) (eff : b Effect.t) -> match eff with | Send n -> Some (fun (k: (b,_) continuation) -> loop_recv n k ()) | Recv -> failwith "protocol violation" | _ -> None } and loop_recv : type a. int -> (a,unit) continuation -> a -> unit = fun n k v -> continue_with k v { retc = Fun.id; exnc = raise; effc = fun (type b) (eff : b Effect.t) -> match eff with | Recv -> Some (fun (k: (b,_) continuation) -> loop_send k n) | Send v -> failwith "protocol violation" | _ -> None } in loop_send (fiber comp) ()

The run function executes the computation comp ensuring that it can only perform an alternating sequence of Send and Recv effects. The shallow handler uses a different set of primitives compared to the deep handler. The primitive fiber (on the last line) takes an 'a -> 'b function and returns a ('a,'b) Effect.Shallow.continuation. The expression continue_with k v h resumes the continuation k with value v under the handler h.

The mutually recursive functions loop_send and loop_recv resume the given continuation k with value v under different handlers. The loop_send function handles the Send effect and tail calls the loop_recv function. If the computation performs the Recv effect, then loop_send aborts the computation by raising an exception. Similarly, the loop_recv function handles the Recv effect and tail calls the loop_send function. If the computation performs the Send effect, then loop_recv aborts the computation. Given that the continuation captured in the shallow handler do not include the handler, there is only ever one handler installed in the dynamic scope of the computation comp.

The computation is initially executed by the loop_send function (see last line in the code above) which ensures that the first effect that the computation is allowed to perform is the Send effect. Note that the computation is free to perform effects other than Send and Recv, which may be handled by an outer handler.

We can see that the run function will permit a computation that follows the protocol:

run (fun () -> printf "Send 42\n"; perform (Send 42); printf "Recv: %d\n" (perform Recv); printf "Send 43\n"; perform (Send 43); printf "Recv: %d\n" (perform Recv))
Send 42 Recv: 42 Send 43 Recv: 43 - : unit = ()

and aborts those that do not:

run (fun () -> Printf.printf "Send 0\n"; perform (Send 0); Printf.printf "Send 1\n"; perform (Send 1) (* protocol violation *))
Send 0 Send 1 Exception: Failure "protocol violation".

We may implement the same example using deep handlers using reference cells (easy, but unsatisfying) or without them (harder). We leave this as an exercise to the reader.

Part III
The OCaml tools

Chapter 13 Batch compilation (ocamlc)

This chapter describes the OCaml batch compiler ocamlc, which compiles OCaml source files to bytecode object files and links these object files to produce standalone bytecode executable files. These executable files are then run by the bytecode interpreter ocamlrun.

13.1 Overview of the compiler

The ocamlc command has a command-line interface similar to the one of most C compilers. It accepts several types of arguments and processes them sequentially, after all options have been processed:

The output of the linking phase is a file containing compiled bytecode that can be executed by the OCaml bytecode interpreter: the command named ocamlrun. If a.out is the name of the file produced by the linking phase, the command

        ocamlrun a.out arg1 arg2 
 argn

executes the compiled code contained in a.out, passing it as arguments the character strings arg1 to argn. (See chapter 15 for more details.)

On most systems, the file produced by the linking phase can be run directly, as in:

        ./a.out arg1 arg2 
 argn

The produced file has the executable bit set, and it manages to launch the bytecode interpreter by itself.

The compiler is able to emit some information on its internal stages. It can output .cmt files for the implementation of the compilation unit and .cmti for signatures if the option -bin-annot is passed to it (see the description of -bin-annot below). Each such file contains a typed abstract syntax tree (AST), that is produced during the type checking procedure. This tree contains all available information about the location and the specific type of each term in the source file. The AST is partial if type checking was unsuccessful.

These .cmt and .cmti files are typically useful for code inspection tools.

13.2 Options

The following command-line options are recognized by ocamlc. The options -pack, -a, -c, -output-obj and -output-complete-obj are mutually exclusive.

-a
Build a library(.cma file) with the object files ( .cmo files) given on the command line, instead of linking them into an executable file. The name of the library must be set with the -o option.

If -custom, -cclib or -ccopt options are passed on the command line, these options are stored in the resulting .cmalibrary. Then, linking with this library automatically adds back the -custom, -cclib and -ccopt options as if they had been provided on the command line, unless the -noautolink option is given.

-absname
Force error messages to show absolute paths for file names.
-no-absname
Do not try to show absolute filenames in error messages.
-annot
Deprecated since OCaml 4.11. Please use -bin-annot instead.
-args filename
Read additional newline-terminated command line arguments from filename.
-args0 filename
Read additional null character terminated command line arguments from filename.
-bin-annot
Dump detailed information about the compilation (types, bindings, tail-calls, etc) in binary format. The information for file src.ml (resp. src.mli) is put into file src.cmt (resp. src.cmti). In case of a type error, dump all the information inferred by the type-checker before the error. The *.cmt and *.cmti files produced by -bin-annot contain more information and are much more compact than the files produced by -annot.
-c
Compile only. Suppress the linking phase of the compilation. Source code files are turned into compiled files, but no executable file is produced. This option is useful to compile modules separately.
-cc ccomp
Use ccomp as the C linker when linking in “custom runtime” mode (see the -custom option) and as the C compiler for compiling .c source files. When linking object files produced by a C++ compiler (such as g++ or clang++), it is recommended to use -cc c++.
-cclib -llibname
Pass the -llibname option to the C linker when linking in “custom runtime” mode (see the -custom option). This causes the given C library to be linked with the program.
-ccopt option
Pass the given option to the C compiler and linker. When linking in “custom runtime” mode, for instance -ccopt -Ldir causes the C linker to search for C libraries in directory dir. (See the -custom option.)
-cmi-file filename
Use the given interface file to type-check the ML source file to compile. When this option is not specified, the compiler looks for a .mli file with the same base name than the implementation it is compiling and in the same directory. If such a file is found, the compiler looks for a corresponding .cmi file in the included directories and reports an error if it fails to find one.
-color mode
Enable or disable colors in compiler messages (especially warnings and errors). The following modes are supported:
auto
use heuristics to enable colors only if the output supports them (an ANSI-compatible tty terminal);
always
enable colors unconditionally;
never
disable color output.

The environment variable OCAML_COLOR is considered if -color is not provided. Its values are auto/always/never as above.

If -color is not provided, OCAML_COLOR is not set and the environment variable NO_COLOR is set, then color output is disabled. Otherwise, the default setting is ’auto’, and the current heuristic checks that the TERM environment variable exists and is not empty or dumb, and that ’isatty(stderr)’ holds.

-error-style mode
Control the way error messages and warnings are printed. The following modes are supported:
short
only print the error and its location;
contextual
like short, but also display the source code snippet corresponding to the location of the error.
The default setting is contextual.

The environment variable OCAML_ERROR_STYLE is considered if -error-style is not provided. Its values are short/contextual as above.

-compat-32
Check that the generated bytecode executable can run on 32-bit platforms and signal an error if it cannot. This is useful when compiling bytecode on a 64-bit machine.
-config
Print the version number of ocamlc and a detailed summary of its configuration, then exit.
-config-var var
Print the value of a specific configuration variable from the -config output, then exit. If the variable does not exist, the exit code is non-zero. This option is only available since OCaml 4.08, so script authors should have a fallback for older versions.
-custom
Link in “custom runtime” mode. In the default linking mode, the linker produces bytecode that is intended to be executed with the shared runtime system, ocamlrun. In the custom runtime mode, the linker produces an output file that contains both the runtime system and the bytecode for the program. The resulting file is larger, but it can be executed directly, even if the ocamlrun command is not installed. Moreover, the “custom runtime” mode enables static linking of OCaml code with user-defined C functions, as described in chapter 22.
Unix:  Never use the strip command on executables produced by ocamlc -custom, this would remove the bytecode part of the executable.
Unix:  Security warning: never set the “setuid” or “setgid” bits on executables produced by ocamlc -custom, this would make them vulnerable to attacks.
-depend ocamldep-args
Compute dependencies, as the ocamldep command would do. The remaining arguments are interpreted as if they were given to the ocamldep command.
-dllib -llibname
Arrange for the C shared library dlllibname.so (dlllibname.dll under Windows) to be loaded dynamically by the run-time system ocamlrun at program start-up time.
-dllpath dir
Adds the directory dir to the run-time search path for shared C libraries. At link-time, shared libraries are searched in the standard search path (the one corresponding to the -I option). The -dllpath option simply stores dir in the produced executable file, where ocamlrun can find it and use it as described in section 15.3.
-for-pack module-path
Generate an object file (.cmo) that can later be included as a sub-module (with the given access path) of a compilation unit constructed with -pack. For instance, ocamlc -for-pack P -c A.ml will generate a..cmo that can later be used with ocamlc -pack -o P.cmo a.cmo. Note: you can still pack a module that was compiled without -for-pack but in this case exceptions will be printed with the wrong names.
-g
Add debugging information while compiling and linking. This option is required in order to be able to debug the program with ocamldebug (see chapter 20), and to produce stack backtraces when the program terminates on an uncaught exception (see section 15.2).
-no-g
Do not record debugging information (default).
-i
Cause the compiler to print all defined names (with their inferred types or their definitions) when compiling an implementation (.ml file). No compiled files (.cmo and .cmi files) are produced. This can be useful to check the types inferred by the compiler. Also, since the output follows the syntax of interfaces, it can help in writing an explicit interface (.mli file) for a file: just redirect the standard output of the compiler to a .mli file, and edit that file to remove all declarations of unexported names.
-I directory
Add the given directory to the list of directories searched for compiled interface files (.cmi), compiled object code files .cmo, libraries (.cma) and C libraries specified with -cclib -lxxx. By default, the current directory is searched first, then the standard library directory. Directories added with -I are searched after the current directory, in the order in which they were given on the command line, but before the standard library directory. See also option -nostdlib.

If the given directory starts with +, it is taken relative to the standard library directory. For instance, -I +unix adds the subdirectory unix of the standard library to the search path.

-impl filename
Compile the file filename as an implementation file, even if its extension is not .ml.
-intf filename
Compile the file filename as an interface file, even if its extension is not .mli.
-intf-suffix string
Recognize file names ending with string as interface files (instead of the default .mli).
-labels
Labels are not ignored in types, labels may be used in applications, and labelled parameters can be given in any order. This is the default.
-linkall
Force all modules contained in libraries to be linked in. If this flag is not given, unreferenced modules are not linked in. When building a library (option -a), setting the -linkall option forces all subsequent links of programs involving that library to link all the modules contained in the library. When compiling a module (option -c), setting the -linkall option ensures that this module will always be linked if it is put in a library and this library is linked.
-make-runtime
Build a custom runtime system (in the file specified by option -o) incorporating the C object files and libraries given on the command line. This custom runtime system can be used later to execute bytecode executables produced with the ocamlc -use-runtime runtime-name option. See section 22.1.6 for more information.
-match-context-rows
Set the number of rows of context used for optimization during pattern matching compilation. The default value is 32. Lower values cause faster compilation, but less optimized code. This advanced option is meant for use in the event that a pattern-match-heavy program leads to significant increases in compilation time.
-no-alias-deps
Do not record dependencies for module aliases. See section 12.8 for more information.
-no-app-funct
Deactivates the applicative behaviour of functors. With this option, each functor application generates new types in its result and applying the same functor twice to the same argument yields two incompatible structures.
-noassert
Do not compile assertion checks. Note that the special form assert false is always compiled because it is typed specially. This flag has no effect when linking already-compiled files.
-noautolink
When linking .cmalibraries, ignore -custom, -cclib and -ccopt options potentially contained in the libraries (if these options were given when building the libraries). This can be useful if a library contains incorrect specifications of C libraries or C options; in this case, during linking, set -noautolink and pass the correct C libraries and options on the command line.
-nolabels
Ignore non-optional labels in types. Labels cannot be used in applications, and parameter order becomes strict.
-nostdlib
Do not include the standard library directory in the list of directories searched for compiled interface files (.cmi), compiled object code files (.cmo), libraries (.cma), and C libraries specified with -cclib -lxxx. See also option -I.
-o output-file
Specify the name of the output file to produce. For executable files, the default output name is a.out under Unix and camlprog.exe under Windows. If the -a option is given, specify the name of the library produced. If the -pack option is given, specify the name of the packed object file produced. If the -output-obj or -output-complete-obj options are given, specify the name of the produced object file. If the -c option is given, specify the name of the object file produced for the next source file that appears on the command line.
-opaque
When the native compiler compiles an implementation, by default it produces a .cmx file containing information for cross-module optimization. It also expects .cmx files to be present for the dependencies of the currently compiled source, and uses them for optimization. Since OCaml 4.03, the compiler will emit a warning if it is unable to locate the .cmx file of one of those dependencies.

The -opaque option, available since 4.04, disables cross-module optimization information for the currently compiled unit. When compiling .mli interface, using -opaque marks the compiled .cmi interface so that subsequent compilations of modules that depend on it will not rely on the corresponding .cmx file, nor warn if it is absent. When the native compiler compiles a .ml implementation, using -opaque generates a .cmx that does not contain any cross-module optimization information.

Using this option may degrade the quality of generated code, but it reduces compilation time, both on clean and incremental builds. Indeed, with the native compiler, when the implementation of a compilation unit changes, all the units that depend on it may need to be recompiled – because the cross-module information may have changed. If the compilation unit whose implementation changed was compiled with -opaque, no such recompilation needs to occur. This option can thus be used, for example, to get faster edit-compile-test feedback loops.

-open Module
Opens the given module before processing the interface or implementation files. If several -open options are given, they are processed in order, just as if the statements open! Module1;; ... open! ModuleN;; were added at the top of each file.
-output-obj
Cause the linker to produce a C object file instead of a bytecode executable file. This is useful to wrap OCaml code as a C library, callable from any C program. See chapter 22, section 22.7.5. The name of the output object file must be set with the -o option. This option can also be used to produce a C source file (.c extension) or a compiled shared/dynamic library (.so extension, .dll under Windows).
-output-complete-exe
Build a self-contained executable by linking a C object file containing the bytecode program, the OCaml runtime system and any other static C code given to ocamlc. The resulting effect is similar to -custom, except that the bytecode is embedded in the C code so it is no longer accessible to tools such as ocamldebug. On the other hand, the resulting binary is resistant to strip.
-output-complete-obj
Same as -output-obj options except the object file produced includes the runtime and autolink libraries.
-pack
Build a bytecode object file (.cmo file) and its associated compiled interface (.cmi) that combines the object files given on the command line, making them appear as sub-modules of the output .cmo file. The name of the output .cmo file must be given with the -o option. For instance,
        ocamlc -pack -o p.cmo a.cmo b.cmo c.cmo
generates compiled files p.cmo and p.cmi describing a compilation unit having three sub-modules A, B and C, corresponding to the contents of the object files a.cmo, b.cmo and c.cmo. These contents can be referenced as P.A, P.B and P.C in the remainder of the program.
-pp command
Cause the compiler to call the given command as a preprocessor for each source file. The output of command is redirected to an intermediate file, which is compiled. If there are no compilation errors, the intermediate file is deleted afterwards.
-ppx command
After parsing, pipe the abstract syntax tree through the preprocessor command. The module Ast_mapper, described in chapter 29: Ast_mapper , implements the external interface of a preprocessor.
-principal
Check information path during type-checking, to make sure that all types are derived in a principal way. When using labelled arguments and/or polymorphic methods, this flag is required to ensure future versions of the compiler will be able to infer types correctly, even if internal algorithms change. All programs accepted in -principal mode are also accepted in the default mode with equivalent types, but different binary signatures, and this may slow down type checking; yet it is a good idea to use it once before publishing source code.
-rectypes
Allow arbitrary recursive types during type-checking. By default, only recursive types where the recursion goes through an object type are supported. Note that once you have created an interface using this flag, you must use it again for all dependencies.
-runtime-variant suffix
Add the suffix string to the name of the runtime library used by the program. Currently, only one such suffix is supported: d, and only if the OCaml compiler was configured with option -with-debug-runtime. This suffix gives the debug version of the runtime, which is useful for debugging pointer problems in low-level code such as C stubs.
-safe-string
Enforce the separation between types string and bytes, thereby making strings read-only. This is the default, and enforced since OCaml 5.0.
-safer-matching
Do not use type information to optimize pattern-matching. This allows to detect match failures even if a pattern-matching was wrongly assumed to be exhaustive. This only impacts GADT and polymorphic variant compilation.
-short-paths
When a type is visible under several module-paths, use the shortest one when printing the type’s name in inferred interfaces and error and warning messages. Identifier names starting with an underscore _ or containing double underscores __ incur a penalty of +10 when computing their length.
-stop-after pass
Stop compilation after the given compilation pass. The currently supported passes are: parsing, typing.
-strict-sequence
Force the left-hand part of each sequence to have type unit.
-strict-formats
Reject invalid formats that were accepted in legacy format implementations. You should use this flag to detect and fix such invalid formats, as they will be rejected by future OCaml versions.
-unboxed-types
When a type is unboxable (i.e. a record with a single argument or a concrete datatype with a single constructor of one argument) it will be unboxed unless annotated with [@@ocaml.boxed].
-no-unboxed-types
When a type is unboxable it will be boxed unless annotated with [@@ocaml.unboxed]. This is the default.
-unsafe
Turn bound checking off for array and string accesses (the v.(i) and s.[i] constructs). Programs compiled with -unsafe are therefore slightly faster, but unsafe: anything can happen if the program accesses an array or string outside of its bounds. Additionally, turn off the check for zero divisor in integer division and modulus operations. With -unsafe, an integer division (or modulus) by zero can halt the program or continue with an unspecified result instead of raising a Division_by_zero exception.
-unsafe-string
Identify the types string and bytes, thereby making strings writable. This is intended for compatibility with old source code and should not be used with new software. This option raises an error unconditionally since OCaml 5.0.
-use-runtime runtime-name
Generate a bytecode executable file that can be executed on the custom runtime system runtime-name, built earlier with ocamlc -make-runtime runtime-name. See section 22.1.6 for more information.
-v
Print the version number of the compiler and the location of the standard library directory, then exit.
-verbose
Print all external commands before they are executed, in particular invocations of the C compiler and linker in -custom mode. Useful to debug C library problems.
-version or -vnum
Print the version number of the compiler in short form (e.g. 3.11.0), then exit.
-w warning-list
Enable, disable, or mark as fatal the warnings specified by the argument warning-list. Each warning can be enabled or disabled, and each warning can be fatal or non-fatal. If a warning is disabled, it isn’t displayed and doesn’t affect compilation in any way (even if it is fatal). If a warning is enabled, it is displayed normally by the compiler whenever the source code triggers it. If it is enabled and fatal, the compiler will also stop with an error after displaying it.

The warning-list argument is a sequence of warning specifiers, with no separators between them. A warning specifier is one of the following:

+num
Enable warning number num.
-num
Disable warning number num.
@num
Enable and mark as fatal warning number num.
+num1..num2
Enable warnings in the given range.
-num1..num2
Disable warnings in the given range.
@num1..num2
Enable and mark as fatal warnings in the given range.
+letter
Enable the set of warnings corresponding to letter. The letter may be uppercase or lowercase.
-letter
Disable the set of warnings corresponding to letter. The letter may be uppercase or lowercase.
@letter
Enable and mark as fatal the set of warnings corresponding to letter. The letter may be uppercase or lowercase.
uppercase-letter
Enable the set of warnings corresponding to uppercase-letter.
lowercase-letter
Disable the set of warnings corresponding to lowercase-letter.

Alternatively, warning-list can specify a single warning using its mnemonic name (see below), as follows:

+name
Enable warning name.
-name
Disable warning name.
@name
Enable and mark as fatal warning name.

Warning numbers, letters and names which are not currently defined are ignored. The warnings are as follows (the name following each number specifies the mnemonic for that warning).

1 comment-start
Suspicious-looking start-of-comment mark.
2 comment-not-end
Suspicious-looking end-of-comment mark.
3
Deprecated synonym for the ’deprecated’ alert.
4 fragile-match
Fragile pattern matching: matching that will remain complete even if additional constructors are added to one of the variant types matched.
5 ignored-partial-application
Partially applied function: expression whose result has function type and is ignored.
6 labels-omitted
Label omitted in function application.
7 method-override
Method overridden.
8 partial-match
Partial match: missing cases in pattern-matching.
9 missing-record-field-pattern
Missing fields in a record pattern.
10 non-unit-statement
Expression on the left-hand side of a sequence that doesn’t have type unit (and that is not a function, see warning number 5).
11 redundant-case
Redundant case in a pattern matching (unused match case).
12 redundant-subpat
Redundant sub-pattern in a pattern-matching.
13 instance-variable-override
Instance variable overridden.
14 illegal-backslash
Illegal backslash escape in a string constant.
15 implicit-public-methods
Private method made public implicitly.
16 unerasable-optional-argument
Unerasable optional argument.
17 undeclared-virtual-method
Undeclared virtual method.
18 not-principal
Non-principal type.
19 non-principal-labels
Type without principality.
20 ignored-extra-argument
Unused function argument.
21 nonreturning-statement
Non-returning statement.
22 preprocessor
Preprocessor warning.
23 useless-record-with
Useless record with clause.
24 bad-module-name
Bad module name: the source file name is not a valid OCaml module name.
25
Ignored: now part of warning 8.
26 unused-var
Suspicious unused variable: unused variable that is bound with let or as, and doesn’t start with an underscore (_) character.
27 unused-var-strict
Innocuous unused variable: unused variable that is not bound with let nor as, and doesn’t start with an underscore (_) character.
28 wildcard-arg-to-constant-constr
Wildcard pattern given as argument to a constant constructor.
29 eol-in-string
Unescaped end-of-line in a string constant (non-portable code).
30 duplicate-definitions
Two labels or constructors of the same name are defined in two mutually recursive types.
31 module-linked-twice
A module is linked twice in the same executable.
I
gnored: now a hard error (since 5.1).
32 unused-value-declaration
Unused value declaration. (since 4.00)
33 unused-open
Unused open statement. (since 4.00)
34 unused-type-declaration
Unused type declaration. (since 4.00)
35 unused-for-index
Unused for-loop index. (since 4.00)
36 unused-ancestor
Unused ancestor variable. (since 4.00)
37 unused-constructor
Unused constructor. (since 4.00)
38 unused-extension
Unused extension constructor. (since 4.00)
39 unused-rec-flag
Unused rec flag. (since 4.00)
40 name-out-of-scope
Constructor or label name used out of scope. (since 4.01)
41 ambiguous-name
Ambiguous constructor or label name. (since 4.01)
42 disambiguated-name
Disambiguated constructor or label name (compatibility warning). (since 4.01)
43 nonoptional-label
Nonoptional label applied as optional. (since 4.01)
44 open-shadow-identifier
Open statement shadows an already defined identifier. (since 4.01)
45 open-shadow-label-constructor
Open statement shadows an already defined label or constructor. (since 4.01)
46 bad-env-variable
Error in environment variable. (since 4.01)
47 attribute-payload
Illegal attribute payload. (since 4.02)
48 eliminated-optional-arguments
Implicit elimination of optional arguments. (since 4.02)
49 no-cmi-file
Absent cmi file when looking up module alias. (since 4.02)
50 unexpected-docstring
Unexpected documentation comment. (since 4.03)
51 wrong-tailcall-expectation
Function call annotated with an incorrect @tailcall attribute. (since 4.03)
52 fragile-literal-pattern (see 13.5.3)
Fragile constant pattern. (since 4.03)
53 misplaced-attribute
Attribute cannot appear in this context. (since 4.03)
54 duplicated-attribute
Attribute used more than once on an expression. (since 4.03)
55 inlining-impossible
Inlining impossible. (since 4.03)
56 unreachable-case
Unreachable case in a pattern-matching (based on type information). (since 4.03)
57 ambiguous-var-in-pattern-guard (see 13.5.4)
Ambiguous or-pattern variables under guard. (since 4.03)
58 no-cmx-file
Missing cmx file. (since 4.03)
59 flambda-assignment-to-non-mutable-value
Assignment to non-mutable value. (since 4.03)
60 unused-module
Unused module declaration. (since 4.04)
61 unboxable-type-in-prim-decl
Unboxable type in primitive declaration. (since 4.04)
62 constraint-on-gadt
Type constraint on GADT type declaration. (since 4.06)
63 erroneous-printed-signature
Erroneous printed signature. (since 4.08)
64 unsafe-array-syntax-without-parsing
-unsafe used with a preprocessor returning a syntax tree. (since 4.08)
65 redefining-unit
Type declaration defining a new ’()’ constructor. (since 4.08)
66 unused-open-bang
Unused open! statement. (since 4.08)
67 unused-functor-parameter
Unused functor parameter. (since 4.10)
68 match-on-mutable-state-prevent-uncurry
Pattern-matching depending on mutable state prevents the remaining arguments from being uncurried. (since 4.12)
69 unused-field
Unused record field. (since 4.13)
70 missing-mli
Missing interface file. (since 4.13)
71 unused-tmc-attribute
Unused @tail_mod_cons attribute. (since 4.14)
72 tmc-breaks-tailcall
A tail call is turned into a non-tail call by the @tail_mod_cons transformation. (since 4.14)
73 generative-application-expects-unit
A generative functor is applied to an empty structure (struct end) rather than to (). (since 5.1)
A
all warnings
C
warnings 1, 2.
D
Alias for warning 3.
E
Alias for warning 4.
F
Alias for warning 5.
K
warnings 32, 33, 34, 35, 36, 37, 38, 39.
L
Alias for warning 6.
M
Alias for warning 7.
P
Alias for warning 8.
R
Alias for warning 9.
S
Alias for warning 10.
U
warnings 11, 12.
V
Alias for warning 13.
X
warnings 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 30.
Y
Alias for warning 26.
Z
Alias for warning 27.

The default setting is -w +a-4-6-7-9-27-29-32..42-44-45-48-50-60. It is displayed by ocamlc -help. Note that warnings 5 and 10 are not always triggered, depending on the internals of the type checker.

-warn-error warning-list
Mark as fatal the warnings specified in the argument warning-list. The compiler will stop with an error when one of these warnings is emitted. The warning-list has the same meaning as for the -w option: a + sign (or an uppercase letter) marks the corresponding warnings as fatal, a - sign (or a lowercase letter) turns them back into non-fatal warnings, and a @ sign both enables and marks as fatal the corresponding warnings.

Note: it is not recommended to use warning sets (i.e. letters) as arguments to -warn-error in production code, because this can break your build when future versions of OCaml add some new warnings.

The default setting is -warn-error -a (no warning is fatal).

-warn-help
Show the description of all available warning numbers.
-where
Print the location of the standard library, then exit.
-with-runtime
Include the runtime system in the generated program. This is the default.
-without-runtime
The compiler does not include the runtime system (nor a reference to it) in the generated program; it must be supplied separately.
- file
Process file as a file name, even if it starts with a dash (-) character.
-help or --help
Display a short usage summary and exit.
contextual-cli-control

Contextual control of command-line options

The compiler command line can be modified “from the outside” with the following mechanisms. These are experimental and subject to change. They should be used only for experimental and development work, not in released packages.

OCAMLPARAM (environment variable)
A set of arguments that will be inserted before or after the arguments from the command line. Arguments are specified in a comma-separated list of name=value pairs. A _ is used to specify the position of the command line arguments, i.e. a=x,_,b=y means that a=x should be executed before parsing the arguments, and b=y after. Finally, an alternative separator can be specified as the first character of the string, within the set :|; ,.
ocaml_compiler_internal_params (file in the stdlib directory)
A mapping of file names to lists of arguments that will be added to the command line (and OCAMLPARAM) arguments.
OCAML_FLEXLINK (environment variable)
Alternative executable to use on native Windows for flexlink instead of the configured value. Primarily used for bootstrapping.

13.3 Modules and the file system

This short section is intended to clarify the relationship between the names of the modules corresponding to compilation units and the names of the files that contain their compiled interface and compiled implementation.

The compiler always derives the module name by taking the capitalized base name of the source file (.ml or .mli file). That is, it strips the leading directory name, if any, as well as the .ml or .mli suffix; then, it set the first letter to uppercase, in order to comply with the requirement that module names must be capitalized. For instance, compiling the file mylib/misc.ml provides an implementation for the module named Misc. Other compilation units may refer to components defined in mylib/misc.ml under the names Misc.name; they can also do open Misc, then use unqualified names name.

The .cmi and .cmo files produced by the compiler have the same base name as the source file. Hence, the compiled files always have their base name equal (modulo capitalization of the first letter) to the name of the module they describe (for .cmi files) or implement (for .cmo files).

When the compiler encounters a reference to a free module identifier Mod, it looks in the search path for a file named Mod.cmi or mod.cmi and loads the compiled interface contained in that file. As a consequence, renaming .cmi files is not advised: the name of a .cmi file must always correspond to the name of the compilation unit it implements. It is admissible to move them to another directory, if their base name is preserved, and the correct -I options are given to the compiler. The compiler will flag an error if it loads a .cmi file that has been renamed.

Compiled bytecode files (.cmo files), on the other hand, can be freely renamed once created. That’s because the linker never attempts to find by itself the .cmo file that implements a module with a given name: it relies instead on the user providing the list of .cmo files by hand.

13.4 Common errors

This section describes and explains the most frequently encountered error messages.

Cannot find file filename
The named file could not be found in the current directory, nor in the directories of the search path. The filename is either a compiled interface file (.cmi file), or a compiled bytecode file (.cmo file). If filename has the format mod.cmi, this means you are trying to compile a file that references identifiers from module mod, but you have not yet compiled an interface for module mod. Fix: compile mod.mli or mod.ml first, to create the compiled interface mod.cmi.

If filename has the format mod.cmo, this means you are trying to link a bytecode object file that does not exist yet. Fix: compile mod.ml first.

If your program spans several directories, this error can also appear because you haven’t specified the directories to look into. Fix: add the correct -I options to the command line.

Corrupted compiled interface filename
The compiler produces this error when it tries to read a compiled interface file (.cmi file) that has the wrong structure. This means something went wrong when this .cmi file was written: the disk was full, the compiler was interrupted in the middle of the file creation, and so on. This error can also appear if a .cmi file is modified after its creation by the compiler. Fix: remove the corrupted .cmi file, and rebuild it.
This expression has type t1, but is used with type t2
This is by far the most common type error in programs. Type t1 is the type inferred for the expression (the part of the program that is displayed in the error message), by looking at the expression itself. Type t2 is the type expected by the context of the expression; it is deduced by looking at how the value of this expression is used in the rest of the program. If the two types t1 and t2 are not compatible, then the error above is produced.

In some cases, it is hard to understand why the two types t1 and t2 are incompatible. For instance, the compiler can report that “expression of type foo cannot be used with type foo”, and it really seems that the two types foo are compatible. This is not always true. Two type constructors can have the same name, but actually represent different types. This can happen if a type constructor is redefined. Example:

        type foo = A | B
        let f = function A -> 0 | B -> 1
        type foo = C | D
        f C

This result in the error message “expression C of type foo cannot be used with type foo”.

The type of this expression, t, contains type variables that cannot be generalized
Type variables ('a, 'b, 
) in a type t can be in either of two states: generalized (which means that the type t is valid for all possible instantiations of the variables) and not generalized (which means that the type t is valid only for one instantiation of the variables). In a let binding let name = expr, the type-checker normally generalizes as many type variables as possible in the type of expr. However, this leads to unsoundness (a well-typed program can crash) in conjunction with polymorphic mutable data structures. To avoid this, generalization is performed at let bindings only if the bound expression expr belongs to the class of “syntactic values”, which includes constants, identifiers, functions, tuples of syntactic values, etc. In all other cases (for instance, expr is a function application), a polymorphic mutable could have been created and generalization is therefore turned off for all variables occurring in contravariant or non-variant branches of the type. For instance, if the type of a non-value is 'a list the variable is generalizable (list is a covariant type constructor), but not in 'a list -> 'a list (the left branch of -> is contravariant) or 'a ref (ref is non-variant).

Non-generalized type variables in a type cause no difficulties inside a given structure or compilation unit (the contents of a .ml file, or an interactive session), but they cannot be allowed inside signatures nor in compiled interfaces (.cmi file), because they could be used inconsistently later. Therefore, the compiler flags an error when a structure or compilation unit defines a value name whose type contains non-generalized type variables. There are two ways to fix this error:

Reference to undefined global mod
This error appears when trying to link an incomplete or incorrectly ordered set of files. Either you have forgotten to provide an implementation for the compilation unit named mod on the command line (typically, the file named mod.cmo, or a library containing that file). Fix: add the missing .ml or .cmo file to the command line. Or, you have provided an implementation for the module named mod, but it comes too late on the command line: the implementation of mod must come before all bytecode object files that reference mod. Fix: change the order of .ml and .cmo files on the command line.

Of course, you will always encounter this error if you have mutually recursive functions across modules. That is, function Mod1.f calls function Mod2.g, and function Mod2.g calls function Mod1.f. In this case, no matter what permutations you perform on the command line, the program will be rejected at link-time. Fixes:

The external function f is not available
This error appears when trying to link code that calls external functions written in C. As explained in chapter 22, such code must be linked with C libraries that implement the required f C function. If the C libraries in question are not shared libraries (DLLs), the code must be linked in “custom runtime” mode. Fix: add the required C libraries to the command line, and possibly the -custom option.

13.5 Warning reference

This section describes and explains in detail some warnings:

13.5.1 Warning 6: Label omitted in function application

OCaml supports labels-omitted full applications: if the function has a known arity, all the arguments are unlabeled, and their number matches the number of non-optional parameters, then labels are ignored and non-optional parameters are matched in their definition order. Optional arguments are defaulted.

let f ~x ~y = x + y
let test = f 2 3

> let test = f 2 3
>            ^
> Warning 6 [labels-omitted]: labels x, y were omitted in the application of this function.

This support for labels-omitted application was introduced when labels were added to OCaml, to ease the progressive introduction of labels in a codebase. However, it has the downside of weakening the labeling discipline: if you use labels to prevent callers from mistakenly reordering two parameters of the same type, labels-omitted make this mistake possible again.

Warning 6 warns when labels-omitted applications are used, to discourage their use. When labels were introduced, this warning was not enabled by default, so users would use labels-omitted applications, often without noticing.

Over time, it has become idiomatic to enable this warning to avoid argument-order mistakes. The warning is now on by default, since OCaml 4.13. Labels-omitted applications are not recommended anymore, but users wishing to preserve this transitory style can disable the warning explicitly.

13.5.2 Warning 9: missing fields in a record pattern

When pattern matching on records, it can be useful to match only few fields of a record. Eliding fields can be done either implicitly or explicitly by ending the record pattern with ; _. However, implicit field elision is at odd with pattern matching exhaustiveness checks. Enabling warning 9 prioritizes exhaustiveness checks over the convenience of implicit field elision and will warn on implicit field elision in record patterns. In particular, this warning can help to spot exhaustive record pattern that may need to be updated after the addition of new fields to a record type.

type 'a point = {x : 'a; y : 'a}
let dx { x } = x (* implicit field elision: trigger warning 9 *)
let dy { y; _ } = y (* explicit field elision: do not trigger warning 9 *)

13.5.3 Warning 52: fragile constant pattern

Some constructors, such as the exception constructors Failure and Invalid_argument, take as parameter a string value holding a text message intended for the user.

These text messages are usually not stable over time: call sites building these constructors may refine the message in a future version to make it more explicit, etc. Therefore, it is dangerous to match over the precise value of the message. For example, until OCaml 4.02, Array.iter2 would raise the exception

  Invalid_argument "arrays must have the same length"

Since 4.03 it raises the more helpful message

  Invalid_argument "Array.iter2: arrays must have the same length"

but this means that any code of the form

  try ...
  with Invalid_argument "arrays must have the same length" -> ...

is now broken and may suffer from uncaught exceptions.

Warning 52 is there to prevent users from writing such fragile code in the first place. It does not occur on every matching on a literal string, but only in the case in which library authors expressed their intent to possibly change the constructor parameter value in the future, by using the attribute ocaml.warn_on_literal_pattern (see the manual section on builtin attributes in 12.12.1):

type t = | Foo of string [@ocaml.warn_on_literal_pattern] | Bar of string let no_warning = function | Bar "specific value" -> 0 | _ -> 1 let warning = function | Foo "specific value" -> 0 | _ -> 1
Warning 52 [fragile-literal-pattern]: Code should not depend on the actual values of this constructor's arguments. They are only for information and may change in future versions. (see manual section 13.5.3)

In particular, all built-in exceptions with a string argument have this attribute set: Invalid_argument, Failure, Sys_error will all raise this warning if you match for a specific string argument.

Additionally, built-in exceptions with a structured argument that includes a string also have the attribute set: Assert_failure and Match_failure will raise the warning for a pattern that uses a literal string to match the first element of their tuple argument.

If your code raises this warning, you should not change the way you test for the specific string to avoid the warning (for example using a string equality inside the right-hand-side instead of a literal pattern), as your code would remain fragile. You should instead enlarge the scope of the pattern by matching on all possible values.

let warning = function
  | Foo _ -> 0
  | _ -> 1

This may require some care: if the scrutinee may return several different cases of the same pattern, or raise distinct instances of the same exception, you may need to modify your code to separate those several cases.

For example,

try (int_of_string count_str, bool_of_string choice_str) with
  | Failure "int_of_string" -> (0, true)
  | Failure "bool_of_string" -> (-1, false)

should be rewritten into more atomic tests. For example, using the exception patterns documented in Section 11.6, one can write:

match int_of_string count_str with
  | exception (Failure _) -> (0, true)
  | count ->
    begin match bool_of_string choice_str with
    | exception (Failure _) -> (-1, false)
    | choice -> (count, choice)
    end

The only case where that transformation is not possible is if a given function call may raise distinct exceptions with the same constructor but different string values. In this case, you will have to check for specific string values. This is dangerous API design and it should be discouraged: it’s better to define more precise exception constructors than store useful information in strings.

13.5.4 Warning 57: Ambiguous or-pattern variables under guard

The semantics of or-patterns in OCaml is specified with a left-to-right bias: a value v matches the pattern p | q if it matches p or q, but if it matches both, the environment captured by the match is the environment captured by p, never the one captured by q.

While this property is generally intuitive, there is at least one specific case where a different semantics might be expected. Consider a pattern followed by a when-guard: | p when g -> e, for example:

     | ((Const x, _) | (_, Const x)) when is_neutral x -> branch

The semantics is clear: match the scrutinee against the pattern, if it matches, test the guard, and if the guard passes, take the branch. In particular, consider the input (Const a, Const b), where a fails the test is_neutral a, while b passes the test is_neutral b. With the left-to-right semantics, the clause above is not taken by its input: matching (Const a, Const b) against the or-pattern succeeds in the left branch, it returns the environment x -> a, and then the guard is_neutral a is tested and fails, the branch is not taken.

However, another semantics may be considered more natural here: any pair that has one side passing the test will take the branch. With this semantics the previous code fragment would be equivalent to

     | (Const x, _) when is_neutral x -> branch
     | (_, Const x) when is_neutral x -> branch

This is not the semantics adopted by OCaml.

Warning 57 is dedicated to these confusing cases where the specified left-to-right semantics is not equivalent to a non-deterministic semantics (any branch can be taken) relatively to a specific guard. More precisely, it warns when guard uses “ambiguous” variables, that are bound to different parts of the scrutinees by different sides of a or-pattern.

Chapter 14 The toplevel system or REPL (ocaml)

This chapter describes the toplevel system for OCaml, that permits interactive use of the OCaml system through a read-eval-print loop (REPL). In this mode, the system repeatedly reads OCaml phrases from the input, then typechecks, compile and evaluate them, then prints the inferred type and result value, if any. The system prints a # (sharp) prompt before reading each phrase.

Input to the toplevel can span several lines. It is terminated by ;; (a double-semicolon). The toplevel input consists in one or several toplevel phrases, with the following syntax:

toplevel-input::= { definition }+ ;;
 âˆŁâ€„expr ;;
 âˆŁâ€„# ident [ directive-argument ] ;;
 
directive-argument::= string-literal
 âˆŁâ€„integer-literal
 âˆŁâ€„value-path
 âˆŁâ€„true ∣ false

A phrase can consist of a definition, like those found in implementations of compilation units or in struct 
 end module expressions. The definition can bind value names, type names, an exception, a module name, or a module type name. The toplevel system performs the bindings, then prints the types and values (if any) for the names thus defined.

A phrase may also consist in a value expression (section 11.7). It is simply evaluated without performing any bindings, and its value is printed.

Finally, a phrase can also consist in a toplevel directive, starting with # (the sharp sign). These directives control the behavior of the toplevel; they are listed below in section 14.2.

Unix:  The toplevel system is started by the command ocaml, as follows:
        ocaml options objects                # interactive mode
        ocaml options objects scriptfile        # script mode
options are described below. objects are filenames ending in .cmo or .cma; they are loaded into the interpreter immediately after options are set. scriptfile is any file name not ending in .cmo or .cma.

If no scriptfile is given on the command line, the toplevel system enters interactive mode: phrases are read on standard input, results are printed on standard output, errors on standard error. End-of-file on standard input terminates ocaml (see also the #quit directive in section 14.2).

On start-up (before the first phrase is read), if the file .ocamlinit exists in the current directory, its contents are read as a sequence of OCaml phrases and executed as per the #use directive described in section 14.2. The evaluation outcode for each phrase are not displayed. If the current directory does not contain an .ocamlinit file, the file XDG_CONFIG_HOME/ocaml/init.ml is looked up according to the XDG base directory specification and used instead (on Windows this is skipped). If that file doesn’t exist then an [.ocamlinit] file in the users’ home directory (determined via environment variable HOME) is used if existing.

The toplevel system does not perform line editing, but it can easily be used in conjunction with an external line editor such as ledit, or rlwrap. An improved toplevel, utop, is also available. Another option is to use ocaml under Gnu Emacs, which gives the full editing power of Emacs (command run-caml from library inf-caml).

At any point, the parsing, compilation or evaluation of the current phrase can be interrupted by pressing ctrl-C (or, more precisely, by sending the INTR signal to the ocaml process). The toplevel then immediately returns to the # prompt.

If scriptfile is given on the command-line to ocaml, the toplevel system enters script mode: the contents of the file are read as a sequence of OCaml phrases and executed, as per the #use directive (section 14.2). The outcome of the evaluation is not printed. On reaching the end of file, the ocaml command exits immediately. No commands are read from standard input. Sys.argv is transformed, ignoring all OCaml parameters, and starting with the script file name in Sys.argv.(0).

In script mode, the first line of the script is ignored if it starts with #!. Thus, it should be possible to make the script itself executable and put as first line #!/usr/local/bin/ocaml, thus calling the toplevel system automatically when the script is run. However, ocaml itself is a #! script on most installations of OCaml, and Unix kernels usually do not handle nested #! scripts. A better solution is to put the following as the first line of the script:

        #!/usr/local/bin/ocamlrun /usr/local/bin/ocaml

14.1 Options

The following command-line options are recognized by the ocaml command.

-absname
Force error messages to show absolute paths for file names.
-no-absname
Do not try to show absolute filenames in error messages.
-args filename
Read additional newline-terminated command line arguments from filename. It is not possible to pass a scriptfile via file to the toplevel.
-args0 filename
Read additional null character terminated command line arguments from filename. It is not possible to pass a scriptfile via file to the toplevel.
-I directory
Add the given directory to the list of directories searched for source and compiled files. By default, the current directory is searched first, then the standard library directory. Directories added with -I are searched after the current directory, in the order in which they were given on the command line, but before the standard library directory. See also option -nostdlib.

If the given directory starts with +, it is taken relative to the standard library directory. For instance, -I +unix adds the subdirectory unix of the standard library to the search path.

Directories can also be added to the list once the toplevel is running with the #directory directive (section 14.2).

-init file
Load the given file instead of the default initialization file. The default file is .ocamlinit in the current directory if it exists, otherwise XDG_CONFIG_HOME/ocaml/init.ml or .ocamlinit in the user’s home directory.
-labels
Labels are not ignored in types, labels may be used in applications, and labelled parameters can be given in any order. This is the default.
-no-app-funct
Deactivates the applicative behaviour of functors. With this option, each functor application generates new types in its result and applying the same functor twice to the same argument yields two incompatible structures.
-noassert
Do not compile assertion checks. Note that the special form assert false is always compiled because it is typed specially.
-nolabels
Ignore non-optional labels in types. Labels cannot be used in applications, and parameter order becomes strict.
-noprompt
Do not display any prompt when waiting for input.
-nopromptcont
Do not display the secondary prompt when waiting for continuation lines in multi-line inputs. This should be used e.g. when running ocaml in an emacs window.
-nostdlib
Do not include the standard library directory in the list of directories searched for source and compiled files.
-ppx command
After parsing, pipe the abstract syntax tree through the preprocessor command. The module Ast_mapper, described in chapter 29: Ast_mapper , implements the external interface of a preprocessor.
-principal
Check information path during type-checking, to make sure that all types are derived in a principal way. When using labelled arguments and/or polymorphic methods, this flag is required to ensure future versions of the compiler will be able to infer types correctly, even if internal algorithms change. All programs accepted in -principal mode are also accepted in the default mode with equivalent types, but different binary signatures, and this may slow down type checking; yet it is a good idea to use it once before publishing source code.
-rectypes
Allow arbitrary recursive types during type-checking. By default, only recursive types where the recursion goes through an object type are supported.
-safe-string
Enforce the separation between types string and bytes, thereby making strings read-only. This is the default, and enforced since OCaml 5.0.
-safer-matching
Do not use type information to optimize pattern-matching. This allows to detect match failures even if a pattern-matching was wrongly assumed to be exhaustive. This only impacts GADT and polymorphic variant compilation.
-short-paths
When a type is visible under several module-paths, use the shortest one when printing the type’s name in inferred interfaces and error and warning messages. Identifier names starting with an underscore _ or containing double underscores __ incur a penalty of +10 when computing their length.
-stdin
Read the standard input as a script file rather than starting an interactive session.
-strict-sequence
Force the left-hand part of each sequence to have type unit.
-strict-formats
Reject invalid formats that were accepted in legacy format implementations. You should use this flag to detect and fix such invalid formats, as they will be rejected by future OCaml versions.
-unsafe
Turn bound checking off for array and string accesses (the v.(i) and s.[i] constructs). Programs compiled with -unsafe are therefore faster, but unsafe: anything can happen if the program accesses an array or string outside of its bounds.
-unsafe-string
Identify the types string and bytes, thereby making strings writable. This is intended for compatibility with old source code and should not be used with new software. This option raises an error unconditionally since OCaml 5.0.
-v
Print the version number of the compiler and the location of the standard library directory, then exit.
-verbose
Print all external commands before they are executed, Useful to debug C library problems.
-version
Print version string and exit.
-vnum
Print short version number and exit.
-no-version
Do not print the version banner at startup.
-w warning-list
Enable, disable, or mark as fatal the warnings specified by the argument warning-list. Each warning can be enabled or disabled, and each warning can be fatal or non-fatal. If a warning is disabled, it isn’t displayed and doesn’t affect compilation in any way (even if it is fatal). If a warning is enabled, it is displayed normally by the compiler whenever the source code triggers it. If it is enabled and fatal, the compiler will also stop with an error after displaying it.

The warning-list argument is a sequence of warning specifiers, with no separators between them. A warning specifier is one of the following:

+num
Enable warning number num.
-num
Disable warning number num.
@num
Enable and mark as fatal warning number num.
+num1..num2
Enable warnings in the given range.
-num1..num2
Disable warnings in the given range.
@num1..num2
Enable and mark as fatal warnings in the given range.
+letter
Enable the set of warnings corresponding to letter. The letter may be uppercase or lowercase.
-letter
Disable the set of warnings corresponding to letter. The letter may be uppercase or lowercase.
@letter
Enable and mark as fatal the set of warnings corresponding to letter. The letter may be uppercase or lowercase.
uppercase-letter
Enable the set of warnings corresponding to uppercase-letter.
lowercase-letter
Disable the set of warnings corresponding to lowercase-letter.

Alternatively, warning-list can specify a single warning using its mnemonic name (see below), as follows:

+name
Enable warning name.
-name
Disable warning name.
@name
Enable and mark as fatal warning name.

Warning numbers, letters and names which are not currently defined are ignored. The warnings are as follows (the name following each number specifies the mnemonic for that warning).

1 comment-start
Suspicious-looking start-of-comment mark.
2 comment-not-end
Suspicious-looking end-of-comment mark.
3
Deprecated synonym for the ’deprecated’ alert.
4 fragile-match
Fragile pattern matching: matching that will remain complete even if additional constructors are added to one of the variant types matched.
5 ignored-partial-application
Partially applied function: expression whose result has function type and is ignored.
6 labels-omitted
Label omitted in function application.
7 method-override
Method overridden.
8 partial-match
Partial match: missing cases in pattern-matching.
9 missing-record-field-pattern
Missing fields in a record pattern.
10 non-unit-statement
Expression on the left-hand side of a sequence that doesn’t have type unit (and that is not a function, see warning number 5).
11 redundant-case
Redundant case in a pattern matching (unused match case).
12 redundant-subpat
Redundant sub-pattern in a pattern-matching.
13 instance-variable-override
Instance variable overridden.
14 illegal-backslash
Illegal backslash escape in a string constant.
15 implicit-public-methods
Private method made public implicitly.
16 unerasable-optional-argument
Unerasable optional argument.
17 undeclared-virtual-method
Undeclared virtual method.
18 not-principal
Non-principal type.
19 non-principal-labels
Type without principality.
20 ignored-extra-argument
Unused function argument.
21 nonreturning-statement
Non-returning statement.
22 preprocessor
Preprocessor warning.
23 useless-record-with
Useless record with clause.
24 bad-module-name
Bad module name: the source file name is not a valid OCaml module name.
25
Ignored: now part of warning 8.
26 unused-var
Suspicious unused variable: unused variable that is bound with let or as, and doesn’t start with an underscore (_) character.
27 unused-var-strict
Innocuous unused variable: unused variable that is not bound with let nor as, and doesn’t start with an underscore (_) character.
28 wildcard-arg-to-constant-constr
Wildcard pattern given as argument to a constant constructor.
29 eol-in-string
Unescaped end-of-line in a string constant (non-portable code).
30 duplicate-definitions
Two labels or constructors of the same name are defined in two mutually recursive types.
31 module-linked-twice
A module is linked twice in the same executable.
I
gnored: now a hard error (since 5.1).
32 unused-value-declaration
Unused value declaration. (since 4.00)
33 unused-open
Unused open statement. (since 4.00)
34 unused-type-declaration
Unused type declaration. (since 4.00)
35 unused-for-index
Unused for-loop index. (since 4.00)
36 unused-ancestor
Unused ancestor variable. (since 4.00)
37 unused-constructor
Unused constructor. (since 4.00)
38 unused-extension
Unused extension constructor. (since 4.00)
39 unused-rec-flag
Unused rec flag. (since 4.00)
40 name-out-of-scope
Constructor or label name used out of scope. (since 4.01)
41 ambiguous-name
Ambiguous constructor or label name. (since 4.01)
42 disambiguated-name
Disambiguated constructor or label name (compatibility warning). (since 4.01)
43 nonoptional-label
Nonoptional label applied as optional. (since 4.01)
44 open-shadow-identifier
Open statement shadows an already defined identifier. (since 4.01)
45 open-shadow-label-constructor
Open statement shadows an already defined label or constructor. (since 4.01)
46 bad-env-variable
Error in environment variable. (since 4.01)
47 attribute-payload
Illegal attribute payload. (since 4.02)
48 eliminated-optional-arguments
Implicit elimination of optional arguments. (since 4.02)
49 no-cmi-file
Absent cmi file when looking up module alias. (since 4.02)
50 unexpected-docstring
Unexpected documentation comment. (since 4.03)
51 wrong-tailcall-expectation
Function call annotated with an incorrect @tailcall attribute. (since 4.03)
52 fragile-literal-pattern (see 13.5.3)
Fragile constant pattern. (since 4.03)
53 misplaced-attribute
Attribute cannot appear in this context. (since 4.03)
54 duplicated-attribute
Attribute used more than once on an expression. (since 4.03)
55 inlining-impossible
Inlining impossible. (since 4.03)
56 unreachable-case
Unreachable case in a pattern-matching (based on type information). (since 4.03)
57 ambiguous-var-in-pattern-guard (see 13.5.4)
Ambiguous or-pattern variables under guard. (since 4.03)
58 no-cmx-file
Missing cmx file. (since 4.03)
59 flambda-assignment-to-non-mutable-value
Assignment to non-mutable value. (since 4.03)
60 unused-module
Unused module declaration. (since 4.04)
61 unboxable-type-in-prim-decl
Unboxable type in primitive declaration. (since 4.04)
62 constraint-on-gadt
Type constraint on GADT type declaration. (since 4.06)
63 erroneous-printed-signature
Erroneous printed signature. (since 4.08)
64 unsafe-array-syntax-without-parsing
-unsafe used with a preprocessor returning a syntax tree. (since 4.08)
65 redefining-unit
Type declaration defining a new ’()’ constructor. (since 4.08)
66 unused-open-bang
Unused open! statement. (since 4.08)
67 unused-functor-parameter
Unused functor parameter. (since 4.10)
68 match-on-mutable-state-prevent-uncurry
Pattern-matching depending on mutable state prevents the remaining arguments from being uncurried. (since 4.12)
69 unused-field
Unused record field. (since 4.13)
70 missing-mli
Missing interface file. (since 4.13)
71 unused-tmc-attribute
Unused @tail_mod_cons attribute. (since 4.14)
72 tmc-breaks-tailcall
A tail call is turned into a non-tail call by the @tail_mod_cons transformation. (since 4.14)
73 generative-application-expects-unit
A generative functor is applied to an empty structure (struct end) rather than to (). (since 5.1)
A
all warnings
C
warnings 1, 2.
D
Alias for warning 3.
E
Alias for warning 4.
F
Alias for warning 5.
K
warnings 32, 33, 34, 35, 36, 37, 38, 39.
L
Alias for warning 6.
M
Alias for warning 7.
P
Alias for warning 8.
R
Alias for warning 9.
S
Alias for warning 10.
U
warnings 11, 12.
V
Alias for warning 13.
X
warnings 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 30.
Y
Alias for warning 26.
Z
Alias for warning 27.

The default setting is -w +a-4-6-7-9-27-29-32..42-44-45-48-50-60. It is displayed by -help. Note that warnings 5 and 10 are not always triggered, depending on the internals of the type checker.

-warn-error warning-list
Mark as fatal the warnings specified in the argument warning-list. The compiler will stop with an error when one of these warnings is emitted. The warning-list has the same meaning as for the -w option: a + sign (or an uppercase letter) marks the corresponding warnings as fatal, a - sign (or a lowercase letter) turns them back into non-fatal warnings, and a @ sign both enables and marks as fatal the corresponding warnings.

Note: it is not recommended to use warning sets (i.e. letters) as arguments to -warn-error in production code, because this can break your build when future versions of OCaml add some new warnings.

The default setting is -warn-error -a (no warning is fatal).

-warn-help
Show the description of all available warning numbers.
- file
Use file as a script file name, even when it starts with a hyphen (-).
-help or --help
Display a short usage summary and exit.
Unix:  The following environment variables are also consulted:
OCAMLTOP_INCLUDE_PATH
Additional directories to search for compiled object code files (.cmi, .cmo and .cma). The specified directories are considered from left to right, after the include directories specified on the command line via -I have been searched. Available since OCaml 4.08.
OCAMLTOP_UTF_8
When printing string values, non-ascii bytes ( > \0x7E ) are printed as decimal escape sequence if OCAMLTOP_UTF_8 is set to false. Otherwise, they are printed unescaped.
TERM
When printing error messages, the toplevel system attempts to underline visually the location of the error. It consults the TERM variable to determines the type of output terminal and look up its capabilities in the terminal database.
XDG_CONFIG_HOME, HOME
.ocamlinit lookup procedure (see above).

14.2 Toplevel directives

The following directives control the toplevel behavior, load files in memory, and trace program execution.

Note: all directives start with a # (sharp) symbol. This # must be typed before the directive, and must not be confused with the # prompt displayed by the interactive loop. For instance, typing #quit;; will exit the toplevel loop, but typing quit;; will result in an “unbound value quit” error.

General
#help;;
Prints a list of all available directives, with corresponding argument type if appropriate.
#quit;;
Exit the toplevel loop and terminate the ocaml command.
Loading codes
#cd "dir-name";;
Change the current working directory.
#directory "dir-name";;
Add the given directory to the list of directories searched for source and compiled files.
#remove_directory "dir-name";;
Remove the given directory from the list of directories searched for source and compiled files. Do nothing if the list does not contain the given directory.
#load "file-name";;
Load in memory a bytecode object file (.cmo file) or library file (.cma file) produced by the batch compiler ocamlc.
#load_rec "file-name";;
Load in memory a bytecode object file (.cmo file) or library file (.cma file) produced by the batch compiler ocamlc. When loading an object file that depends on other modules which have not been loaded yet, the .cmo files for these modules are searched and loaded as well, recursively. The loading order is not specified.
#use "file-name";;
Read, compile and execute source phrases from the given file. This is textual inclusion: phrases are processed just as if they were typed on standard input. The reading of the file stops at the first error encountered.
#use_output "command";;
Execute a command and evaluate its output as if it had been captured to a file and passed to #use.
#mod_use "file-name";;
Similar to #use but also wrap the code into a top-level module of the same name as capitalized file name without extensions, following semantics of the compiler.

For directives that take file names as arguments, if the given file name specifies no directory, the file is searched in the following directories:

  1. In script mode, the directory containing the script currently executing; in interactive mode, the current working directory.
  2. Directories added with the #directory directive.
  3. Directories given on the command line with -I options.
  4. The standard library directory.
Environment queries
#show_class class-path;;
#show_class_type class-path;;
#show_exception ident;;
#show_module module-path;;
#show_module_type modtype-path;;
#show_type typeconstr;;
#show_val value-path;;
Print the signature of the corresponding component.
#show ident;;
Print the signatures of components with name ident in all the above categories.
Pretty-printing
#install_printer printer-name;;
This directive registers the function named printer-name (a value path) as a printer for values whose types match the argument type of the function. That is, the toplevel loop will call printer-name when it has such a value to print.

The printing function printer-name should have type Format.formatter -> t -> unit, where t is the type for the values to be printed, and should output its textual representation for the value of type t on the given formatter, using the functions provided by the Format library. For backward compatibility, printer-name can also have type t -> unit and should then output on the standard formatter, but this usage is deprecated.

#print_depth n;;
Limit the printing of values to a maximal depth of n. The parts of values whose depth exceeds n are printed as ... (ellipsis).
#print_length n;;
Limit the number of value nodes printed to at most n. Remaining parts of values are printed as ... (ellipsis).
#remove_printer printer-name;;
Remove the named function from the table of toplevel printers.
Tracing
#trace function-name;;
After executing this directive, all calls to the function named function-name will be “traced”. That is, the argument and the result are displayed for each call, as well as the exceptions escaping out of the function, raised either by the function itself or by another function it calls. If the function is curried, each argument is printed as it is passed to the function.
#untrace function-name;;
Stop tracing the given function.
#untrace_all;;
Stop tracing all functions traced so far.
Compiler options
#debug bool;;
Turn on/off the insertion of debugging events. Default is true.
#labels bool;;
Ignore labels in function types if argument is false, or switch back to default behaviour (commuting style) if argument is true.
#ppx "file-name";;
After parsing, pipe the abstract syntax tree through the preprocessor command.
#principal bool;;
If the argument is true, check information paths during type-checking, to make sure that all types are derived in a principal way. If the argument is false, do not check information paths.
#rectypes;;
Allow arbitrary recursive types during type-checking. Note: once enabled, this option cannot be disabled because that would lead to unsoundness of the type system.
#warn_error "warning-list";;
Treat as errors the warnings enabled by the argument and as normal warnings the warnings disabled by the argument.
#warnings "warning-list";;
Enable or disable warnings according to the argument.

14.3 The toplevel and the module system

Toplevel phrases can refer to identifiers defined in compilation units with the same mechanisms as for separately compiled units: either by using qualified names (Modulename.localname), or by using the open construct and unqualified names (see section 11.3).

However, before referencing another compilation unit, an implementation of that unit must be present in memory. At start-up, the toplevel system contains implementations for all the modules in the the standard library. Implementations for user modules can be entered with the #load directive described above. Referencing a unit for which no implementation has been provided results in the error Reference to undefined global `...'.

Note that entering open Mod merely accesses the compiled interface (.cmi file) for Mod, but does not load the implementation of Mod, and does not cause any error if no implementation of Mod has been loaded. The error “reference to undefined global Mod” will occur only when executing a value or module definition that refers to Mod.

14.4 Common errors

This section describes and explains the most frequently encountered error messages.

Cannot find file filename
The named file could not be found in the current directory, nor in the directories of the search path.

If filename has the format mod.cmi, this means you have referenced the compilation unit mod, but its compiled interface could not be found. Fix: compile mod.mli or mod.ml first, to create the compiled interface mod.cmi.

If filename has the format mod.cmo, this means you are trying to load with #load a bytecode object file that does not exist yet. Fix: compile mod.ml first.

If your program spans several directories, this error can also appear because you haven’t specified the directories to look into. Fix: use the #directory directive to add the correct directories to the search path.

This expression has type t1, but is used with type t2
See section 13.4.
Reference to undefined global mod
You have neglected to load in memory an implementation for a module with #load. See section 14.3 above.

14.5 Building custom toplevel systems: ocamlmktop

The ocamlmktop command builds OCaml toplevels that contain user code preloaded at start-up.

The ocamlmktop command takes as argument a set of .cmo and .cma files, and links them with the object files that implement the OCaml toplevel. The typical use is:

        ocamlmktop -o mytoplevel foo.cmo bar.cmo gee.cmo

This creates the bytecode file mytoplevel, containing the OCaml toplevel system, plus the code from the three .cmo files. This toplevel is directly executable and is started by:

        ./mytoplevel

This enters a regular toplevel loop, except that the code from foo.cmo, bar.cmo and gee.cmo is already loaded in memory, just as if you had typed:

        #load "foo.cmo";;
        #load "bar.cmo";;
        #load "gee.cmo";;

on entrance to the toplevel. The modules Foo, Bar and Gee are not opened, though; you still have to do

        open Foo;;

yourself, if this is what you wish.

14.5.1 Options

The following command-line options are recognized by ocamlmktop.

-cclib libname
Pass the -llibname option to the C linker when linking in “custom runtime” mode. See the corresponding option for ocamlc, in chapter 13.
-ccopt option
Pass the given option to the C compiler and linker, when linking in “custom runtime” mode. See the corresponding option for ocamlc, in chapter 13.
-custom
Link in “custom runtime” mode. See the corresponding option for ocamlc, in chapter 13.
-I directory
Add the given directory to the list of directories searched for compiled object code files (.cmo and .cma).
-o exec-file
Specify the name of the toplevel file produced by the linker. The default is a.out.

14.6 The native toplevel: ocamlnat (experimental)

This section describes a tool that is not yet officially supported but may be found useful.

OCaml code executing in the traditional toplevel system uses the bytecode interpreter. When increased performance is required, or for testing programs that will only execute correctly when compiled to native code, the native toplevel may be used instead.

For the majority of installations the native toplevel will not have been installed along with the rest of the OCaml toolchain. In such circumstances it will be necessary to build the OCaml distribution from source. From the built source tree of the distribution you may use make natruntop to build and execute a native toplevel. (Alternatively make ocamlnat can be used, which just performs the build step.)

If the make install command is run after having built the native toplevel then the ocamlnat program (either from the source or the installation directory) may be invoked directly rather than using make natruntop.

Chapter 15 The runtime system (ocamlrun)

The ocamlrun command executes bytecode files produced by the linking phase of the ocamlc command.

15.1 Overview

The ocamlrun command comprises three main parts: the bytecode interpreter, that actually executes bytecode files; the memory allocator and garbage collector; and a set of C functions that implement primitive operations such as input/output.

The usage for ocamlrun is:

        ocamlrun options bytecode-executable arg1 ... argn

The first non-option argument is taken to be the name of the file containing the executable bytecode. (That file is searched in the executable path as well as in the current directory.) The remaining arguments are passed to the OCaml program, in the string array Sys.argv. Element 0 of this array is the name of the bytecode executable file; elements 1 to n are the remaining arguments arg1 to argn.

As mentioned in chapter 13, the bytecode executable files produced by the ocamlc command are self-executable, and manage to launch the ocamlrun command on themselves automatically. That is, assuming a.out is a bytecode executable file,

        a.out arg1 ... argn

works exactly as

        ocamlrun a.out arg1 ... argn

Notice that it is not possible to pass options to ocamlrun when invoking a.out directly.

Windows:  Under several versions of Windows, bytecode executable files are self-executable only if their name ends in .exe. It is recommended to always give .exe names to bytecode executables, e.g. compile with ocamlc -o myprog.exe ... rather than ocamlc -o myprog ....

15.2 Options

The following command-line options are recognized by ocamlrun.

-b
When the program aborts due to an uncaught exception, print a detailed “back trace” of the execution, showing where the exception was raised and which function calls were outstanding at this point. The back trace is printed only if the bytecode executable contains debugging information, i.e. was compiled and linked with the -g option to ocamlc set. This is equivalent to setting the b flag in the OCAMLRUNPARAM environment variable (see below).
-config
Print the version number of ocamlrun and a detailed summary of its configuration, then exit.
-I dir
Search the directory dir for dynamically-loaded libraries, in addition to the standard search path (see section 15.3).
-m
Print the magic number of the bytecode executable given as argument and exit.
-M
Print the magic number expected for bytecode executables by this version of the runtime and exit.
-p
Print the names of the primitives known to this version of ocamlrun and exit.
-t
Increments the trace level for the debug runtime (ignored otherwise).
-v
Direct the memory manager to print some progress messages on standard error. This is equivalent to setting v=61 in the OCAMLRUNPARAM environment variable (see below).
-version
Print version string and exit.
-vnum
Print short version number and exit.

The following environment variables are also consulted:

CAML_LD_LIBRARY_PATH
Additional directories to search for dynamically-loaded libraries (see section 15.3).
OCAMLLIB
The directory containing the OCaml standard library. (If OCAMLLIB is not set, CAMLLIB will be used instead.) Used to locate the ld.conf configuration file for dynamic loading (see section 15.3). If not set, default to the library directory specified when compiling OCaml.
OCAMLRUNPARAM
Set the runtime system options and garbage collection parameters. (If OCAMLRUNPARAM is not set, CAMLRUNPARAM will be used instead.) This variable must be a sequence of parameter specifications separated by commas. For convenience, commas at the beginning of the variable are ignored, and multiple runs of commas are interpreted as a single one. A parameter specification is an option letter followed by an = sign, a decimal number (or an hexadecimal number prefixed by 0x), and an optional multiplier. The options are documented below; the options a, i, l, m, M, n, o, O, s, v, w correspond to the fields of the control record documented in Module Gc.
b
(backtrace) Trigger the printing of a stack backtrace when an uncaught exception aborts the program. An optional argument can be provided: b=0 turns backtrace printing off; b=1 is equivalent to b and turns backtrace printing on; b=2 turns backtrace printing on and forces the runtime system to load debugging information at program startup time instead of at backtrace printing time. b=2 can be used if the runtime is unable to load debugging information at backtrace printing time, for example if there are no file descriptors available.
c
(cleanup_on_exit) Shut the runtime down gracefully on exit (see caml_shutdown in section 22.7.5). The option also enables pooling (as in caml_startup_pooled). This mode can be used to detect leaks with a third-party memory debugger.
e
(runtime_events_log_wsize) Size of the per-domain runtime events ring buffers in log powers of two words. Defaults to 16, giving 64k word or 512kb buffers on 64-bit systems.
l
(stack_limit) The limit (in words) of the stack size. This is only relevant to the byte-code runtime, as the native code runtime uses the operating system’s stack.
m
(custom_minor_ratio) Bound on floating garbage for out-of-heap memory held by custom values in the minor heap. A minor GC is triggered when this much memory is held by custom values located in the minor heap. Expressed as a percentage of minor heap size. Default: 100. Note: this only applies to values allocated with caml_alloc_custom_mem.
M
(custom_major_ratio) Target ratio of floating garbage to major heap size for out-of-heap memory held by custom values (e.g. bigarrays) located in the major heap. The GC speed is adjusted to try to use this much memory for dead values that are not yet collected. Expressed as a percentage of major heap size. Default: 44. Note: this only applies to values allocated with caml_alloc_custom_mem.
n
(custom_minor_max_size) Maximum amount of out-of-heap memory for each custom value allocated in the minor heap. When a custom value is allocated on the minor heap and holds more than this many bytes, only this value is counted against custom_minor_ratio and the rest is directly counted against custom_major_ratio. Default: 8192 bytes. Note: this only applies to values allocated with caml_alloc_custom_mem.
The multiplier is k, M, or G, for multiplication by 210, 220, and 230 respectively.
o
(space_overhead) The major GC speed setting. See the Gc module documentation for details.
p
(parser trace) Turn on debugging support for ocamlyacc-generated parsers. When this option is on, the pushdown automaton that executes the parsers prints a trace of its actions. This option takes no argument.
R
(randomize) Turn on randomization of all hash tables by default (see Module Hashtbl). This option takes no argument.
s
(minor_heap_size) Size of the minor heap. (in words)
t
Set the trace level for the debug runtime (ignored by the standard runtime).
v
(verbose) What GC messages to print to stderr. This is a sum of values selected from the following:
1 (= 0x001)
Start and end of major GC cycle.
2 (= 0x002)
Minor collection and major GC slice.
4 (= 0x004)
Growing and shrinking of the heap.
8 (= 0x008)
Resizing of stacks and memory manager tables.
16 (= 0x010)
Heap compaction.
32 (= 0x020)
Change of GC parameters.
64 (= 0x040)
Computation of major GC slice size.
128 (= 0x080)
Calling of finalization functions
256 (= 0x100)
Startup messages (loading the bytecode executable file, resolving shared libraries).
512 (= 0x200)
Computation of compaction-triggering condition.
1024 (= 0x400)
Output GC statistics at program exit.
2048 (= 0x800)
GC debugging messages.
4096 (= 0x1000)
Address space reservation changes.
V
(verify_heap) runs an integrity check on the heap just after the completion of a major GC cycle
W
Print runtime warnings to stderr (such as Channel opened on file dies without being closed, unflushed data, etc.)

If the option letter is not recognized, the whole parameter is ignored; if the equal sign or the number is missing, the value is taken as 1; if the multiplier is not recognized, it is ignored.

For example, on a 32-bit machine, under bash the command

        export OCAMLRUNPARAM='b,s=256k,v=0x015'

tells a subsequent ocamlrun to print backtraces for uncaught exceptions, set its initial minor heap size to 1 megabyte and print a message at the start of each major GC cycle, when the heap size changes, and when compaction is triggered.

CAMLRUNPARAM
If OCAMLRUNPARAM is not found in the environment, then CAMLRUNPARAM will be used instead. If CAMLRUNPARAM is also not found, then the default values will be used.
PATH
List of directories searched to find the bytecode executable file.

15.3 Dynamic loading of shared libraries

On platforms that support dynamic loading, ocamlrun can link dynamically with C shared libraries (DLLs) providing additional C primitives beyond those provided by the standard runtime system. The names for these libraries are provided at link time as described in section 22.1.4), and recorded in the bytecode executable file; ocamlrun, then, locates these libraries and resolves references to their primitives when the bytecode executable program starts.

The ocamlrun command searches shared libraries in the following directories, in the order indicated:

  1. Directories specified on the ocamlrun command line with the -I option.
  2. Directories specified in the CAML_LD_LIBRARY_PATH environment variable.
  3. Directories specified at link-time via the -dllpath option to ocamlc. (These directories are recorded in the bytecode executable file.)
  4. Directories specified in the file ld.conf. This file resides in the OCaml standard library directory, and lists directory names (one per line) to be searched. Typically, it contains only one line naming the stublibs subdirectory of the OCaml standard library directory. Users can add there the names of other directories containing frequently-used shared libraries; however, for consistency of installation, we recommend that shared libraries are installed directly in the system stublibs directory, rather than adding lines to the ld.conf file.
  5. Default directories searched by the system dynamic loader. Under Unix, these generally include /lib and /usr/lib, plus the directories listed in the file /etc/ld.so.conf and the environment variable LD_LIBRARY_PATH. Under Windows, these include the Windows system directories, plus the directories listed in the PATH environment variable.

15.4 Common errors

This section describes and explains the most frequently encountered error messages.

filename: no such file or directory
If filename is the name of a self-executable bytecode file, this means that either that file does not exist, or that it failed to run the ocamlrun bytecode interpreter on itself. The second possibility indicates that OCaml has not been properly installed on your system.
Cannot exec ocamlrun
(When launching a self-executable bytecode file.) The ocamlrun could not be found in the executable path. Check that OCaml has been properly installed on your system.
Cannot find the bytecode file
The file that ocamlrun is trying to execute (e.g. the file given as first non-option argument to ocamlrun) either does not exist, or is not a valid executable bytecode file.
Truncated bytecode file
The file that ocamlrun is trying to execute is not a valid executable bytecode file. Probably it has been truncated or mangled since created. Erase and rebuild it.
Uncaught exception
The program being executed contains a “stray” exception. That is, it raises an exception at some point, and this exception is never caught. This causes immediate termination of the program. The name of the exception is printed, along with its string, byte sequence, and integer arguments (arguments of more complex types are not correctly printed). To locate the context of the uncaught exception, compile the program with the -g option and either run it again under the ocamldebug debugger (see chapter 20), or run it with ocamlrun -b or with the OCAMLRUNPARAM environment variable set to b=1.
Out of memory
The program being executed requires more memory than available. Either the program builds excessively large data structures; or the program contains too many nested function calls, and the stack overflows. In some cases, your program is perfectly correct, it just requires more memory than your machine provides. In other cases, the “out of memory” message reveals an error in your program: non-terminating recursive function, allocation of an excessively large array, string or byte sequence, attempts to build an infinite list or other data structure, 


To help you diagnose this error, run your program with the -v option to ocamlrun, or with the OCAMLRUNPARAM environment variable set to v=63. If it displays lots of “Growing stack
” messages, this is probably a looping recursive function. If it displays lots of “Growing heap
” messages, with the heap size growing slowly, this is probably an attempt to construct a data structure with too many (infinitely many?) cells. If it displays few “Growing heap
” messages, but with a huge increment in the heap size, this is probably an attempt to build an excessively large array, string or byte sequence.

Chapter 16 Native-code compilation (ocamlopt)

This chapter describes the OCaml high-performance native-code compiler ocamlopt, which compiles OCaml source files to native code object files and links these object files to produce standalone executables.

The native-code compiler is only available on certain platforms. It produces code that runs faster than the bytecode produced by ocamlc, at the cost of increased compilation time and executable code size. Compatibility with the bytecode compiler is extremely high: the same source code should run identically when compiled with ocamlc and ocamlopt.

It is not possible to mix native-code object files produced by ocamlopt with bytecode object files produced by ocamlc: a program must be compiled entirely with ocamlopt or entirely with ocamlc. Native-code object files produced by ocamlopt cannot be loaded in the toplevel system ocaml.

16.1 Overview of the compiler

The ocamlopt command has a command-line interface very close to that of ocamlc. It accepts the same types of arguments, and processes them sequentially, after all options have been processed:

The output of the linking phase is a regular Unix or Windows executable file. It does not need ocamlrun to run.

The compiler is able to emit some information on its internal stages:

16.2 Options

The following command-line options are recognized by ocamlopt. The options -pack, -a, -shared, -c, -output-obj and -output-complete-obj are mutually exclusive.

-a
Build a library(.cmxa and .a/.lib files) with the object files (.cmx and .o/.obj files) given on the command line, instead of linking them into an executable file. The name of the library must be set with the -o option.

If -cclib or -ccopt options are passed on the command line, these options are stored in the resulting .cmxalibrary. Then, linking with this library automatically adds back the -cclib and -ccopt options as if they had been provided on the command line, unless the -noautolink option is given.

-absname
Force error messages to show absolute paths for file names.
-no-absname
Do not try to show absolute filenames in error messages.
-annot
Deprecated since OCaml 4.11. Please use -bin-annot instead.
-args filename
Read additional newline-terminated command line arguments from filename.
-args0 filename
Read additional null character terminated command line arguments from filename.
-bin-annot
Dump detailed information about the compilation (types, bindings, tail-calls, etc) in binary format. The information for file src.ml (resp. src.mli) is put into file src.cmt (resp. src.cmti). In case of a type error, dump all the information inferred by the type-checker before the error. The *.cmt and *.cmti files produced by -bin-annot contain more information and are much more compact than the files produced by -annot.
-c
Compile only. Suppress the linking phase of the compilation. Source code files are turned into compiled files, but no executable file is produced. This option is useful to compile modules separately.
-cc ccomp
Use ccomp as the C linker called to build the final executable and as the C compiler for compiling .c source files. When linking object files produced by a C++ compiler (such as g++ or clang++), it is recommended to use -cc c++.
-cclib -llibname
Pass the -llibname option to the linker . This causes the given C library to be linked with the program.
-ccopt option
Pass the given option to the C compiler and linker. For instance, -ccopt -Ldir causes the C linker to search for C libraries in directory dir.
-cmi-file filename
Use the given interface file to type-check the ML source file to compile. When this option is not specified, the compiler looks for a .mli file with the same base name than the implementation it is compiling and in the same directory. If such a file is found, the compiler looks for a corresponding .cmi file in the included directories and reports an error if it fails to find one.
-color mode
Enable or disable colors in compiler messages (especially warnings and errors). The following modes are supported:
auto
use heuristics to enable colors only if the output supports them (an ANSI-compatible tty terminal);
always
enable colors unconditionally;
never
disable color output.

The environment variable OCAML_COLOR is considered if -color is not provided. Its values are auto/always/never as above.

If -color is not provided, OCAML_COLOR is not set and the environment variable NO_COLOR is set, then color output is disabled. Otherwise, the default setting is ’auto’, and the current heuristic checks that the TERM environment variable exists and is not empty or dumb, and that ’isatty(stderr)’ holds.

-error-style mode
Control the way error messages and warnings are printed. The following modes are supported:
short
only print the error and its location;
contextual
like short, but also display the source code snippet corresponding to the location of the error.
The default setting is contextual.

The environment variable OCAML_ERROR_STYLE is considered if -error-style is not provided. Its values are short/contextual as above.

-compact
Optimize the produced code for space rather than for time. This results in slightly smaller but slightly slower programs. The default is to optimize for speed.
-config
Print the version number of ocamlopt and a detailed summary of its configuration, then exit.
-config-var var
Print the value of a specific configuration variable from the -config output, then exit. If the variable does not exist, the exit code is non-zero. This option is only available since OCaml 4.08, so script authors should have a fallback for older versions.
-depend ocamldep-args
Compute dependencies, as the ocamldep command would do. The remaining arguments are interpreted as if they were given to the ocamldep command.
-for-pack module-path
Generate an object file (.cmx and .o/.obj files) that can later be included as a sub-module (with the given access path) of a compilation unit constructed with -pack. For instance, ocamlopt -for-pack P -c A.ml will generate a..cmx and a.o files that can later be used with ocamlopt -pack -o P.cmx a.cmx. Note: you can still pack a module that was compiled without -for-pack but in this case exceptions will be printed with the wrong names.
-g
Add debugging information while compiling and linking. This option is required in order to produce stack backtraces when the program terminates on an uncaught exception (see section 15.2).
-no-g
Do not record debugging information (default).
-i
Cause the compiler to print all defined names (with their inferred types or their definitions) when compiling an implementation (.ml file). No compiled files (.cmo and .cmi files) are produced. This can be useful to check the types inferred by the compiler. Also, since the output follows the syntax of interfaces, it can help in writing an explicit interface (.mli file) for a file: just redirect the standard output of the compiler to a .mli file, and edit that file to remove all declarations of unexported names.
-I directory
Add the given directory to the list of directories searched for compiled interface files (.cmi), compiled object code files (.cmx), and libraries (.cmxa). By default, the current directory is searched first, then the standard library directory. Directories added with -I are searched after the current directory, in the order in which they were given on the command line, but before the standard library directory. See also option -nostdlib.

If the given directory starts with +, it is taken relative to the standard library directory. For instance, -I +unix adds the subdirectory unix of the standard library to the search path.

-impl filename
Compile the file filename as an implementation file, even if its extension is not .ml.
-inline n
Set aggressiveness of inlining to n, where n is a positive integer. Specifying -inline 0 prevents all functions from being inlined, except those whose body is smaller than the call site. Thus, inlining causes no expansion in code size. The default aggressiveness, -inline 1, allows slightly larger functions to be inlined, resulting in a slight expansion in code size. Higher values for the -inline option cause larger and larger functions to become candidate for inlining, but can result in a serious increase in code size.
-intf filename
Compile the file filename as an interface file, even if its extension is not .mli.
-intf-suffix string
Recognize file names ending with string as interface files (instead of the default .mli).
-labels
Labels are not ignored in types, labels may be used in applications, and labelled parameters can be given in any order. This is the default.
-linkall
Force all modules contained in libraries to be linked in. If this flag is not given, unreferenced modules are not linked in. When building a library (option -a), setting the -linkall option forces all subsequent links of programs involving that library to link all the modules contained in the library. When compiling a module (option -c), setting the -linkall option ensures that this module will always be linked if it is put in a library and this library is linked.
-linscan
Use linear scan register allocation. Compiling with this allocator is faster than with the usual graph coloring allocator, sometimes quite drastically so for long functions and modules. On the other hand, the generated code can be a bit slower.
-match-context-rows
Set the number of rows of context used for optimization during pattern matching compilation. The default value is 32. Lower values cause faster compilation, but less optimized code. This advanced option is meant for use in the event that a pattern-match-heavy program leads to significant increases in compilation time.
-no-alias-deps
Do not record dependencies for module aliases. See section 12.8 for more information.
-no-app-funct
Deactivates the applicative behaviour of functors. With this option, each functor application generates new types in its result and applying the same functor twice to the same argument yields two incompatible structures.
-no-float-const-prop
Deactivates the constant propagation for floating-point operations. This option should be given if the program changes the float rounding mode during its execution.
-noassert
Do not compile assertion checks. Note that the special form assert false is always compiled because it is typed specially. This flag has no effect when linking already-compiled files.
-noautolink
When linking .cmxalibraries, ignore -cclib and -ccopt options potentially contained in the libraries (if these options were given when building the libraries). This can be useful if a library contains incorrect specifications of C libraries or C options; in this case, during linking, set -noautolink and pass the correct C libraries and options on the command line.
-nodynlink
Allow the compiler to use some optimizations that are valid only for code that is statically linked to produce a non-relocatable executable. The generated code cannot be linked to produce a shared library nor a position-independent executable (PIE). Many operating systems produce PIEs by default, causing errors when linking code compiled with -nodynlink. Either do not use -nodynlink or pass the option -ccopt -no-pie at link-time.
-nolabels
Ignore non-optional labels in types. Labels cannot be used in applications, and parameter order becomes strict.
-nostdlib
Do not automatically add the standard library directory to the list of directories searched for compiled interface files (.cmi), compiled object code files (.cmx), and libraries (.cmxa). See also option -I.
-o output-file
Specify the name of the output file to produce. For executable files, the default output name is a.out under Unix and camlprog.exe under Windows. If the -a option is given, specify the name of the library produced. If the -pack option is given, specify the name of the packed object file produced. If the -output-obj or -output-complete-obj options are given, specify the name of the produced object file. If the -shared option is given, specify the name of plugin file produced.
-opaque
When the native compiler compiles an implementation, by default it produces a .cmx file containing information for cross-module optimization. It also expects .cmx files to be present for the dependencies of the currently compiled source, and uses them for optimization. Since OCaml 4.03, the compiler will emit a warning if it is unable to locate the .cmx file of one of those dependencies.

The -opaque option, available since 4.04, disables cross-module optimization information for the currently compiled unit. When compiling .mli interface, using -opaque marks the compiled .cmi interface so that subsequent compilations of modules that depend on it will not rely on the corresponding .cmx file, nor warn if it is absent. When the native compiler compiles a .ml implementation, using -opaque generates a .cmx that does not contain any cross-module optimization information.

Using this option may degrade the quality of generated code, but it reduces compilation time, both on clean and incremental builds. Indeed, with the native compiler, when the implementation of a compilation unit changes, all the units that depend on it may need to be recompiled – because the cross-module information may have changed. If the compilation unit whose implementation changed was compiled with -opaque, no such recompilation needs to occur. This option can thus be used, for example, to get faster edit-compile-test feedback loops.

-open Module
Opens the given module before processing the interface or implementation files. If several -open options are given, they are processed in order, just as if the statements open! Module1;; ... open! ModuleN;; were added at the top of each file.
-output-obj
Cause the linker to produce a C object file instead of an executable file. This is useful to wrap OCaml code as a C library, callable from any C program. See chapter 22, section 22.7.5. The name of the output object file must be set with the -o option. This option can also be used to produce a compiled shared/dynamic library (.so extension, .dll under Windows).
-output-complete-obj
Same as -output-obj options except the object file produced includes the runtime and autolink libraries.
-pack
Build an object file (.cmx and .o/.obj files) and its associated compiled interface (.cmi) that combines the .cmx object files given on the command line, making them appear as sub-modules of the output .cmx file. The name of the output .cmx file must be given with the -o option. For instance,
        ocamlopt -pack -o P.cmx A.cmx B.cmx C.cmx
generates compiled files P.cmx, P.o and P.cmi describing a compilation unit having three sub-modules A, B and C, corresponding to the contents of the object files A.cmx, B.cmx and C.cmx. These contents can be referenced as P.A, P.B and P.C in the remainder of the program.

The .cmx object files being combined must have been compiled with the appropriate -for-pack option. In the example above, A.cmx, B.cmx and C.cmx must have been compiled with ocamlopt -for-pack P.

Multiple levels of packing can be achieved by combining -pack with -for-pack. Consider the following example:

        ocamlopt -for-pack P.Q -c A.ml
        ocamlopt -pack -o Q.cmx -for-pack P A.cmx
        ocamlopt -for-pack P -c B.ml
        ocamlopt -pack -o P.cmx Q.cmx B.cmx

The resulting P.cmx object file has sub-modules P.Q, P.Q.A and P.B.

-pp command
Cause the compiler to call the given command as a preprocessor for each source file. The output of command is redirected to an intermediate file, which is compiled. If there are no compilation errors, the intermediate file is deleted afterwards.
-ppx command
After parsing, pipe the abstract syntax tree through the preprocessor command. The module Ast_mapper, described in chapter 29: Ast_mapper , implements the external interface of a preprocessor.
-principal
Check information path during type-checking, to make sure that all types are derived in a principal way. When using labelled arguments and/or polymorphic methods, this flag is required to ensure future versions of the compiler will be able to infer types correctly, even if internal algorithms change. All programs accepted in -principal mode are also accepted in the default mode with equivalent types, but different binary signatures, and this may slow down type checking; yet it is a good idea to use it once before publishing source code.
-rectypes
Allow arbitrary recursive types during type-checking. By default, only recursive types where the recursion goes through an object type are supported. Note that once you have created an interface using this flag, you must use it again for all dependencies.
-runtime-variant suffix
Add the suffix string to the name of the runtime library used by the program. Currently, only one such suffix is supported: d, and only if the OCaml compiler was configured with option -with-debug-runtime. This suffix gives the debug version of the runtime, which is useful for debugging pointer problems in low-level code such as C stubs.
-S
Keep the assembly code produced during the compilation. The assembly code for the source file x.ml is saved in the file x.s.
-safe-string
Enforce the separation between types string and bytes, thereby making strings read-only. This is the default, and enforced since OCaml 5.0.
-safer-matching
Do not use type information to optimize pattern-matching. This allows to detect match failures even if a pattern-matching was wrongly assumed to be exhaustive. This only impacts GADT and polymorphic variant compilation.
-save-ir-after pass
Save intermediate representation after the given compilation pass to a file. The currently supported passes and the corresponding file extensions are: scheduling (.cmir-linear).

This experimental feature enables external tools to inspect and manipulate compiler’s intermediate representation of the program using compiler-libs library (see chapter 29 and Compiler_libs ).

-shared
Build a plugin (usually .cmxs) that can be dynamically loaded with the Dynlink module. The name of the plugin must be set with the -o option. A plugin can include a number of OCaml modules and libraries, and extra native objects (.o, .obj, .a, .lib files). Building native plugins is only supported for some operating system. Under some systems (currently, only Linux AMD 64), all the OCaml code linked in a plugin must have been compiled without the -nodynlink flag. Some constraints might also apply to the way the extra native objects have been compiled (under Linux AMD 64, they must contain only position-independent code).
-short-paths
When a type is visible under several module-paths, use the shortest one when printing the type’s name in inferred interfaces and error and warning messages. Identifier names starting with an underscore _ or containing double underscores __ incur a penalty of +10 when computing their length.
-stop-after pass
Stop compilation after the given compilation pass. The currently supported passes are: parsing, typing, scheduling, emit.
-strict-sequence
Force the left-hand part of each sequence to have type unit.
-strict-formats
Reject invalid formats that were accepted in legacy format implementations. You should use this flag to detect and fix such invalid formats, as they will be rejected by future OCaml versions.
-unboxed-types
When a type is unboxable (i.e. a record with a single argument or a concrete datatype with a single constructor of one argument) it will be unboxed unless annotated with [@@ocaml.boxed].
-no-unboxed-types
When a type is unboxable it will be boxed unless annotated with [@@ocaml.unboxed]. This is the default.
-unsafe
Turn bound checking off for array and string accesses (the v.(i) and s.[i] constructs). Programs compiled with -unsafe are therefore faster, but unsafe: anything can happen if the program accesses an array or string outside of its bounds. Additionally, turn off the check for zero divisor in integer division and modulus operations. With -unsafe, an integer division (or modulus) by zero can halt the program or continue with an unspecified result instead of raising a Division_by_zero exception.
-unsafe-string
Identify the types string and bytes, thereby making strings writable. This is intended for compatibility with old source code and should not be used with new software. This option raises an error unconditionally since OCaml 5.0.
-v
Print the version number of the compiler and the location of the standard library directory, then exit.
-verbose
Print all external commands before they are executed, in particular invocations of the assembler, C compiler, and linker. Useful to debug C library problems.
-version or -vnum
Print the version number of the compiler in short form (e.g. 3.11.0), then exit.
-w warning-list
Enable, disable, or mark as fatal the warnings specified by the argument warning-list. Each warning can be enabled or disabled, and each warning can be fatal or non-fatal. If a warning is disabled, it isn’t displayed and doesn’t affect compilation in any way (even if it is fatal). If a warning is enabled, it is displayed normally by the compiler whenever the source code triggers it. If it is enabled and fatal, the compiler will also stop with an error after displaying it.

The warning-list argument is a sequence of warning specifiers, with no separators between them. A warning specifier is one of the following:

+num
Enable warning number num.
-num
Disable warning number num.
@num
Enable and mark as fatal warning number num.
+num1..num2
Enable warnings in the given range.
-num1..num2
Disable warnings in the given range.
@num1..num2
Enable and mark as fatal warnings in the given range.
+letter
Enable the set of warnings corresponding to letter. The letter may be uppercase or lowercase.
-letter
Disable the set of warnings corresponding to letter. The letter may be uppercase or lowercase.
@letter
Enable and mark as fatal the set of warnings corresponding to letter. The letter may be uppercase or lowercase.
uppercase-letter
Enable the set of warnings corresponding to uppercase-letter.
lowercase-letter
Disable the set of warnings corresponding to lowercase-letter.

Alternatively, warning-list can specify a single warning using its mnemonic name (see below), as follows:

+name
Enable warning name.
-name
Disable warning name.
@name
Enable and mark as fatal warning name.

Warning numbers, letters and names which are not currently defined are ignored. The warnings are as follows (the name following each number specifies the mnemonic for that warning).

1 comment-start
Suspicious-looking start-of-comment mark.
2 comment-not-end
Suspicious-looking end-of-comment mark.
3
Deprecated synonym for the ’deprecated’ alert.
4 fragile-match
Fragile pattern matching: matching that will remain complete even if additional constructors are added to one of the variant types matched.
5 ignored-partial-application
Partially applied function: expression whose result has function type and is ignored.
6 labels-omitted
Label omitted in function application.
7 method-override
Method overridden.
8 partial-match
Partial match: missing cases in pattern-matching.
9 missing-record-field-pattern
Missing fields in a record pattern.
10 non-unit-statement
Expression on the left-hand side of a sequence that doesn’t have type unit (and that is not a function, see warning number 5).
11 redundant-case
Redundant case in a pattern matching (unused match case).
12 redundant-subpat
Redundant sub-pattern in a pattern-matching.
13 instance-variable-override
Instance variable overridden.
14 illegal-backslash
Illegal backslash escape in a string constant.
15 implicit-public-methods
Private method made public implicitly.
16 unerasable-optional-argument
Unerasable optional argument.
17 undeclared-virtual-method
Undeclared virtual method.
18 not-principal
Non-principal type.
19 non-principal-labels
Type without principality.
20 ignored-extra-argument
Unused function argument.
21 nonreturning-statement
Non-returning statement.
22 preprocessor
Preprocessor warning.
23 useless-record-with
Useless record with clause.
24 bad-module-name
Bad module name: the source file name is not a valid OCaml module name.
25
Ignored: now part of warning 8.
26 unused-var
Suspicious unused variable: unused variable that is bound with let or as, and doesn’t start with an underscore (_) character.
27 unused-var-strict
Innocuous unused variable: unused variable that is not bound with let nor as, and doesn’t start with an underscore (_) character.
28 wildcard-arg-to-constant-constr
Wildcard pattern given as argument to a constant constructor.
29 eol-in-string
Unescaped end-of-line in a string constant (non-portable code).
30 duplicate-definitions
Two labels or constructors of the same name are defined in two mutually recursive types.
31 module-linked-twice
A module is linked twice in the same executable.
I
gnored: now a hard error (since 5.1).
32 unused-value-declaration
Unused value declaration. (since 4.00)
33 unused-open
Unused open statement. (since 4.00)
34 unused-type-declaration
Unused type declaration. (since 4.00)
35 unused-for-index
Unused for-loop index. (since 4.00)
36 unused-ancestor
Unused ancestor variable. (since 4.00)
37 unused-constructor
Unused constructor. (since 4.00)
38 unused-extension
Unused extension constructor. (since 4.00)
39 unused-rec-flag
Unused rec flag. (since 4.00)
40 name-out-of-scope
Constructor or label name used out of scope. (since 4.01)
41 ambiguous-name
Ambiguous constructor or label name. (since 4.01)
42 disambiguated-name
Disambiguated constructor or label name (compatibility warning). (since 4.01)
43 nonoptional-label
Nonoptional label applied as optional. (since 4.01)
44 open-shadow-identifier
Open statement shadows an already defined identifier. (since 4.01)
45 open-shadow-label-constructor
Open statement shadows an already defined label or constructor. (since 4.01)
46 bad-env-variable
Error in environment variable. (since 4.01)
47 attribute-payload
Illegal attribute payload. (since 4.02)
48 eliminated-optional-arguments
Implicit elimination of optional arguments. (since 4.02)
49 no-cmi-file
Absent cmi file when looking up module alias. (since 4.02)
50 unexpected-docstring
Unexpected documentation comment. (since 4.03)
51 wrong-tailcall-expectation
Function call annotated with an incorrect @tailcall attribute. (since 4.03)
52 fragile-literal-pattern (see 13.5.3)
Fragile constant pattern. (since 4.03)
53 misplaced-attribute
Attribute cannot appear in this context. (since 4.03)
54 duplicated-attribute
Attribute used more than once on an expression. (since 4.03)
55 inlining-impossible
Inlining impossible. (since 4.03)
56 unreachable-case
Unreachable case in a pattern-matching (based on type information). (since 4.03)
57 ambiguous-var-in-pattern-guard (see 13.5.4)
Ambiguous or-pattern variables under guard. (since 4.03)
58 no-cmx-file
Missing cmx file. (since 4.03)
59 flambda-assignment-to-non-mutable-value
Assignment to non-mutable value. (since 4.03)
60 unused-module
Unused module declaration. (since 4.04)
61 unboxable-type-in-prim-decl
Unboxable type in primitive declaration. (since 4.04)
62 constraint-on-gadt
Type constraint on GADT type declaration. (since 4.06)
63 erroneous-printed-signature
Erroneous printed signature. (since 4.08)
64 unsafe-array-syntax-without-parsing
-unsafe used with a preprocessor returning a syntax tree. (since 4.08)
65 redefining-unit
Type declaration defining a new ’()’ constructor. (since 4.08)
66 unused-open-bang
Unused open! statement. (since 4.08)
67 unused-functor-parameter
Unused functor parameter. (since 4.10)
68 match-on-mutable-state-prevent-uncurry
Pattern-matching depending on mutable state prevents the remaining arguments from being uncurried. (since 4.12)
69 unused-field
Unused record field. (since 4.13)
70 missing-mli
Missing interface file. (since 4.13)
71 unused-tmc-attribute
Unused @tail_mod_cons attribute. (since 4.14)
72 tmc-breaks-tailcall
A tail call is turned into a non-tail call by the @tail_mod_cons transformation. (since 4.14)
73 generative-application-expects-unit
A generative functor is applied to an empty structure (struct end) rather than to (). (since 5.1)
A
all warnings
C
warnings 1, 2.
D
Alias for warning 3.
E
Alias for warning 4.
F
Alias for warning 5.
K
warnings 32, 33, 34, 35, 36, 37, 38, 39.
L
Alias for warning 6.
M
Alias for warning 7.
P
Alias for warning 8.
R
Alias for warning 9.
S
Alias for warning 10.
U
warnings 11, 12.
V
Alias for warning 13.
X
warnings 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 30.
Y
Alias for warning 26.
Z
Alias for warning 27.

The default setting is -w +a-4-6-7-9-27-29-32..42-44-45-48-50-60. It is displayed by ocamlopt -help. Note that warnings 5 and 10 are not always triggered, depending on the internals of the type checker.

-warn-error warning-list
Mark as fatal the warnings specified in the argument warning-list. The compiler will stop with an error when one of these warnings is emitted. The warning-list has the same meaning as for the -w option: a + sign (or an uppercase letter) marks the corresponding warnings as fatal, a - sign (or a lowercase letter) turns them back into non-fatal warnings, and a @ sign both enables and marks as fatal the corresponding warnings.

Note: it is not recommended to use warning sets (i.e. letters) as arguments to -warn-error in production code, because this can break your build when future versions of OCaml add some new warnings.

The default setting is -warn-error -a (no warning is fatal).

-warn-help
Show the description of all available warning numbers.
-where
Print the location of the standard library, then exit.
-with-runtime
Include the runtime system in the generated program. This is the default.
-without-runtime
The compiler does not include the runtime system (nor a reference to it) in the generated program; it must be supplied separately.
- file
Process file as a file name, even if it starts with a dash (-) character.
-help or --help
Display a short usage summary and exit.
Options for the 64-bit x86 architecture

The 64-bit code generator for Intel/AMD x86 processors (amd64 architecture) supports the following additional options:

-fPIC
Generate position-independent machine code. This is the default.
-fno-PIC
Generate position-dependent machine code.
Options for the PowerPC architecture

The PowerPC code generator supports the following additional options:

-flarge-toc
Enables the PowerPC large model allowing the TOC (table of contents) to be arbitrarily large. This is the default since 4.11.
-fsmall-toc
Enables the PowerPC small model allowing the TOC to be up to 64 kbytes per compilation unit. Prior to 4.11 this was the default behaviour.
Contextual control of command-line options

The compiler command line can be modified “from the outside” with the following mechanisms. These are experimental and subject to change. They should be used only for experimental and development work, not in released packages.

OCAMLPARAM (environment variable)
A set of arguments that will be inserted before or after the arguments from the command line. Arguments are specified in a comma-separated list of name=value pairs. A _ is used to specify the position of the command line arguments, i.e. a=x,_,b=y means that a=x should be executed before parsing the arguments, and b=y after. Finally, an alternative separator can be specified as the first character of the string, within the set :|; ,.
ocaml_compiler_internal_params (file in the stdlib directory)
A mapping of file names to lists of arguments that will be added to the command line (and OCAMLPARAM) arguments.
OCAML_FLEXLINK (environment variable)
Alternative executable to use on native Windows for flexlink instead of the configured value. Primarily used for bootstrapping.

16.3 Common errors

The error messages are almost identical to those of ocamlc. See section 13.4.

16.4 Running executables produced by ocamlopt

Executables generated by ocamlopt are native, stand-alone executable files that can be invoked directly. They do not depend on the ocamlrun bytecode runtime system nor on dynamically-loaded C/OCaml stub libraries.

During execution of an ocamlopt-generated executable, the following environment variables are also consulted:

OCAMLRUNPARAM
Same usage as in ocamlrun (see section 15.2), except that option l is ignored (the operating system’s stack size limit is used instead).
CAMLRUNPARAM
If OCAMLRUNPARAM is not found in the environment, then CAMLRUNPARAM will be used instead. If CAMLRUNPARAM is not found, then the default values will be used.

16.5 Compatibility with the bytecode compiler

This section lists the known incompatibilities between the bytecode compiler and the native-code compiler. Except on those points, the two compilers should generate code that behave identically.

Chapter 17 Lexer and parser generators (ocamllex, ocamlyacc)

This chapter describes two program generators: ocamllex, that produces a lexical analyzer from a set of regular expressions with associated semantic actions, and ocamlyacc, that produces a parser from a grammar with associated semantic actions.

These program generators are very close to the well-known lex and yacc commands that can be found in most C programming environments. This chapter assumes a working knowledge of lex and yacc: while it describes the input syntax for ocamllex and ocamlyacc and the main differences with lex and yacc, it does not explain the basics of writing a lexer or parser description in lex and yacc. Readers unfamiliar with lex and yacc are referred to “Compilers: principles, techniques, and tools” by Aho, Lam, Sethi and Ullman (Pearson, 2006), or “Lex & Yacc”, by Levine, Mason and Brown (O’Reilly, 1992).

17.1 Overview of ocamllex

The ocamllex command produces a lexical analyzer from a set of regular expressions with attached semantic actions, in the style of lex. Assuming the input file is lexer.mll, executing

        ocamllex lexer.mll

produces OCaml code for a lexical analyzer in file lexer.ml. This file defines one lexing function per entry point in the lexer definition. These functions have the same names as the entry points. Lexing functions take as argument a lexer buffer, and return the semantic attribute of the corresponding entry point.

Lexer buffers are an abstract data type implemented in the standard library module Lexing. The functions Lexing.from_channel, Lexing.from_string and Lexing.from_function create lexer buffers that read from an input channel, a character string, or any reading function, respectively. (See the description of module Lexing in chapter 28.)

When used in conjunction with a parser generated by ocamlyacc, the semantic actions compute a value belonging to the type token defined by the generated parsing module. (See the description of ocamlyacc below.)

17.1.1 Options

The following command-line options are recognized by ocamllex.

-ml
Output code that does not use OCaml’s built-in automata interpreter. Instead, the automaton is encoded by OCaml functions. This option improves performance when using the native compiler, but decreases it when using the bytecode compiler.
-o output-file
Specify the name of the output file produced by ocamllex. The default is the input file name with its extension replaced by .ml.
-q
Quiet mode. ocamllex normally outputs informational messages to standard output. They are suppressed if option -q is used.
-v or -version
Print version string and exit.
-vnum
Print short version number and exit.
-help or --help
Display a short usage summary and exit.

17.2 Syntax of lexer definitions

The format of lexer definitions is as follows:

{ header }
let ident = regexp 

[refill { refill-handler }]
rule entrypoint [arg1
 argn] =
  parse regexp { action }
      | 

      | regexp { action }
and entrypoint [arg1
 argn] =
  parse 

and 

{ trailer }

Comments are delimited by (* and *), as in OCaml. The parse keyword, can be replaced by the shortest keyword, with the semantic consequences explained below.

Refill handlers are a recent (optional) feature introduced in 4.02, documented below in subsection 17.2.7.

17.2.1 Header and trailer

The header and trailer sections are arbitrary OCaml text enclosed in curly braces. Either or both can be omitted. If present, the header text is copied as is at the beginning of the output file and the trailer text at the end. Typically, the header section contains the open directives required by the actions, and possibly some auxiliary functions used in the actions.

17.2.2 Naming regular expressions

Between the header and the entry points, one can give names to frequently-occurring regular expressions. This is written let ident = regexp. In regular expressions that follow this declaration, the identifier ident can be used as shorthand for regexp.

17.2.3 Entry points

The names of the entry points must be valid identifiers for OCaml values (starting with a lowercase letter). Similarly, the arguments arg1
 argn must be valid identifiers for OCaml. Each entry point becomes an OCaml function that takes n+1 arguments, the extra implicit last argument being of type Lexing.lexbuf. Characters are read from the Lexing.lexbuf argument and matched against the regular expressions provided in the rule, until a prefix of the input matches one of the rule. The corresponding action is then evaluated and returned as the result of the function.

If several regular expressions match a prefix of the input, the “longest match” rule applies: the regular expression that matches the longest prefix of the input is selected. In case of tie, the regular expression that occurs earlier in the rule is selected.

However, if lexer rules are introduced with the shortest keyword in place of the parse keyword, then the “shortest match” rule applies: the shortest prefix of the input is selected. In case of tie, the regular expression that occurs earlier in the rule is still selected. This feature is not intended for use in ordinary lexical analyzers, it may facilitate the use of ocamllex as a simple text processing tool.

17.2.4 Regular expressions

The regular expressions are in the style of lex, with a more OCaml-like syntax.

regexp::= 

' regular-char ∣ escape-sequence '
A character constant, with the same syntax as OCaml character constants. Match the denoted character.
_
(underscore) Match any character.
eof
Match the end of the lexer input.
Note: On some systems, with interactive input, an end-of-file may be followed by more characters. However, ocamllex will not correctly handle regular expressions that contain eof followed by something else.
" { string-character } "
A string constant, with the same syntax as OCaml string constants. Match the corresponding sequence of characters.
[ character-set ]
Match any single character belonging to the given character set. Valid character sets are: single character constants ' c '; ranges of characters ' c1 ' - ' c2 ' (all characters between c1 and c2, inclusive); and the union of two or more character sets, denoted by concatenation.
[ ^ character-set ]
Match any single character not belonging to the given character set.
regexp1 # regexp2
(difference of character sets) Regular expressions regexp1 and regexp2 must be character sets defined with [
 ] (or a single character expression or underscore _). Match the difference of the two specified character sets.
regexp *
(repetition) Match the concatenation of zero or more strings that match regexp.
regexp +
(strict repetition) Match the concatenation of one or more strings that match regexp.
regexp ?
(option) Match the empty string, or a string matching regexp.
regexp1 | regexp2
(alternative) Match any string that matches regexp1 or regexp2. If both regexp1 and regexp2 are character sets, this constructions produces another character set, obtained by taking the union of regexp1 and regexp2.
regexp1 regexp2
(concatenation) Match the concatenation of two strings, the first matching regexp1, the second matching regexp2.
( regexp )
Match the same strings as regexp.
ident
Reference the regular expression bound to ident by an earlier let ident = regexp definition.
regexp as ident
Bind the substring matched by regexp to identifier ident.

Concerning the precedences of operators, # has the highest precedence, followed by *, + and ?, then concatenation, then | (alternation), then as.

17.2.5 Actions

The actions are arbitrary OCaml expressions. They are evaluated in a context where the identifiers defined by using the as construct are bound to subparts of the matched string. Additionally, lexbuf is bound to the current lexer buffer. Some typical uses for lexbuf, in conjunction with the operations on lexer buffers provided by the Lexing standard library module, are listed below.

Lexing.lexeme lexbuf
Return the matched string.
Lexing.lexeme_char lexbuf n
Return the nth character in the matched string. The first character corresponds to n = 0.
Lexing.lexeme_start lexbuf
Return the absolute position in the input text of the beginning of the matched string (i.e. the offset of the first character of the matched string). The first character read from the input text has offset 0.
Lexing.lexeme_end lexbuf
Return the absolute position in the input text of the end of the matched string (i.e. the offset of the first character after the matched string). The first character read from the input text has offset 0.
entrypoint [exp1
 expn] lexbuf
(Where entrypoint is the name of another entry point in the same lexer definition.) Recursively call the lexer on the given entry point. Notice that lexbuf is the last argument. Useful for lexing nested comments, for example.

17.2.6 Variables in regular expressions

The as construct is similar to “groups” as provided by numerous regular expression packages. The type of these variables can be string, char, string option or char option.

We first consider the case of linear patterns, that is the case when all as bound variables are distinct. In regexp as ident, the type of ident normally is string (or string option) except when regexp is a character constant, an underscore, a string constant of length one, a character set specification, or an alternation of those. Then, the type of ident is char (or char option). Option types are introduced when overall rule matching does not imply matching of the bound sub-pattern. This is in particular the case of ( regexp as ident ) ? and of regexp1 | ( regexp2 as ident ).

There is no linearity restriction over as bound variables. When a variable is bound more than once, the previous rules are to be extended as follows:

For instance, in ('a' as x) | ( 'a' (_ as x) ) the variable x is of type char, whereas in ("ab" as x) | ( 'a' (_ as x) ? ) the variable x is of type string option.

In some cases, a successful match may not yield a unique set of bindings. For instance the matching of aba by the regular expression (('a'|"ab") as x) (("ba"|'a') as y) may result in binding either x to "ab" and y to "a", or x to "a" and y to "ba". The automata produced ocamllex on such ambiguous regular expressions will select one of the possible resulting sets of bindings. The selected set of bindings is purposely left unspecified.

17.2.7 Refill handlers

By default, when ocamllex reaches the end of its lexing buffer, it will silently call the refill_buff function of lexbuf structure and continue lexing. It is sometimes useful to be able to take control of refilling action; typically, if you use a library for asynchronous computation, you may want to wrap the refilling action in a delaying function to avoid blocking synchronous operations.

Since OCaml 4.02, it is possible to specify a refill-handler, a function that will be called when refill happens. It is passed the continuation of the lexing, on which it has total control. The OCaml expression used as refill action should have a type that is an instance of

   (Lexing.lexbuf -> 'a) -> Lexing.lexbuf -> 'a

where the first argument is the continuation which captures the processing ocamllex would usually perform (refilling the buffer, then calling the lexing function again), and the result type that instantiates [’a] should unify with the result type of all lexing rules.

As an example, consider the following lexer that is parametrized over an arbitrary monad:

{
type token = EOL | INT of int | PLUS

module Make (M : sig
               type 'a t
               val return: 'a -> 'a t
               val bind: 'a t -> ('a -> 'b t) -> 'b t
               val fail : string -> 'a t

               (* Set up lexbuf *)
               val on_refill : Lexing.lexbuf -> unit t
             end)
= struct

let refill_handler k lexbuf =
    M.bind (M.on_refill lexbuf) (fun () -> k lexbuf)

}

refill {refill_handler}

rule token = parse
| [' ' '\t']
    { token lexbuf }
| '\n'
    { M.return EOL }
| ['0'-'9']+ as i
    { M.return (INT (int_of_string i)) }
| '+'
    { M.return PLUS }
| _
    { M.fail "unexpected character" }
{
end
}

17.2.8 Reserved identifiers

All identifiers starting with __ocaml_lex are reserved for use by ocamllex; do not use any such identifier in your programs.

17.3 Overview of ocamlyacc

The ocamlyacc command produces a parser from a context-free grammar specification with attached semantic actions, in the style of yacc. Assuming the input file is grammar.mly, executing

        ocamlyacc options grammar.mly

produces OCaml code for a parser in the file grammar.ml, and its interface in file grammar.mli.

The generated module defines one parsing function per entry point in the grammar. These functions have the same names as the entry points. Parsing functions take as arguments a lexical analyzer (a function from lexer buffers to tokens) and a lexer buffer, and return the semantic attribute of the corresponding entry point. Lexical analyzer functions are usually generated from a lexer specification by the ocamllex program. Lexer buffers are an abstract data type implemented in the standard library module Lexing. Tokens are values from the concrete type token, defined in the interface file grammar.mli produced by ocamlyacc.

17.4 Syntax of grammar definitions

Grammar definitions have the following format:

%{
  header
%}
  declarations
%%
  rules
%%
  trailer

Comments are delimited by (* and *), as in OCaml. Additionally, comments can be delimited by /* and */, as in C, in the “declarations” and “rules” sections. C-style comments do not nest, but OCaml-style comments do.

17.4.1 Header and trailer

The header and the trailer sections are OCaml code that is copied as is into file grammar.ml. Both sections are optional. The header goes at the beginning of the output file; it usually contains open directives and auxiliary functions required by the semantic actions of the rules. The trailer goes at the end of the output file.

17.4.2 Declarations

Declarations are given one per line. They all start with a % sign.

%token constr 
 constr
Declare the given symbols constr 
 constr as tokens (terminal symbols). These symbols are added as constant constructors for the token concrete type.
%token < typexpr > constr 
 constr
Declare the given symbols constr 
 constr as tokens with an attached attribute of the given type. These symbols are added as constructors with arguments of the given type for the token concrete type. The typexpr part is an arbitrary OCaml type expression, except that all type constructor names must be fully qualified (e.g. Modname.typename) for all types except standard built-in types, even if the proper open directives (e.g. open Modname) were given in the header section. That’s because the header is copied only to the .ml output file, but not to the .mli output file, while the typexpr part of a %token declaration is copied to both.
%start symbol 
 symbol
Declare the given symbols as entry points for the grammar. For each entry point, a parsing function with the same name is defined in the output module. Non-terminals that are not declared as entry points have no such parsing function. Start symbols must be given a type with the %type directive below.
%type < typexpr > symbol 
 symbol
Specify the type of the semantic attributes for the given symbols. This is mandatory for start symbols only. Other nonterminal symbols need not be given types by hand: these types will be inferred when running the output files through the OCaml compiler (unless the -s option is in effect). The typexpr part is an arbitrary OCaml type expression, except that all type constructor names must be fully qualified, as explained above for %token.
%left symbol 
 symbol
%right symbol 
 symbol
%nonassoc symbol 
 symbol

Associate precedences and associativities to the given symbols. All symbols on the same line are given the same precedence. They have higher precedence than symbols declared before in a %left, %right or %nonassoc line. They have lower precedence than symbols declared after in a %left, %right or %nonassoc line. The symbols are declared to associate to the left (%left), to the right (%right), or to be non-associative (%nonassoc). The symbols are usually tokens. They can also be dummy nonterminals, for use with the %prec directive inside the rules.

The precedence declarations are used in the following way to resolve reduce/reduce and shift/reduce conflicts:

17.4.3 Rules

The syntax for rules is as usual:

nonterminal :
    symbol 
 symbol { semantic-action }
  | 

  | symbol 
 symbol { semantic-action }
;

Rules can also contain the %prec symbol directive in the right-hand side part, to override the default precedence and associativity of the rule with the precedence and associativity of the given symbol.

Semantic actions are arbitrary OCaml expressions, that are evaluated to produce the semantic attribute attached to the defined nonterminal. The semantic actions can access the semantic attributes of the symbols in the right-hand side of the rule with the $ notation: $1 is the attribute for the first (leftmost) symbol, $2 is the attribute for the second symbol, etc.

The rules may contain the special symbol error to indicate resynchronization points, as in yacc.

Actions occurring in the middle of rules are not supported.

Nonterminal symbols are like regular OCaml symbols, except that they cannot end with ' (single quote).

17.4.4 Error handling

Error recovery is supported as follows: when the parser reaches an error state (no grammar rules can apply), it calls a function named parse_error with the string "syntax error" as argument. The default parse_error function does nothing and returns, thus initiating error recovery (see below). The user can define a customized parse_error function in the header section of the grammar file.

The parser also enters error recovery mode if one of the grammar actions raises the Parsing.Parse_error exception.

In error recovery mode, the parser discards states from the stack until it reaches a place where the error token can be shifted. It then discards tokens from the input until it finds three successive tokens that can be accepted, and starts processing with the first of these. If no state can be uncovered where the error token can be shifted, then the parser aborts by raising the Parsing.Parse_error exception.

Refer to documentation on yacc for more details and guidance in how to use error recovery.

17.5 Options

The ocamlyacc command recognizes the following options:

-bprefix
Name the output files prefix.ml, prefix.mli, prefix.output, instead of the default naming convention.
-q
This option has no effect.
-v
Generate a description of the parsing tables and a report on conflicts resulting from ambiguities in the grammar. The description is put in file grammar.output.
-version
Print version string and exit.
-vnum
Print short version number and exit.
-
Read the grammar specification from standard input. The default output file names are stdin.ml and stdin.mli.
-- file
Process file as the grammar specification, even if its name starts with a dash (-) character. This option must be the last on the command line.

At run-time, the ocamlyacc-generated parser can be debugged by setting the p option in the OCAMLRUNPARAM environment variable (see section 15.2). This causes the pushdown automaton executing the parser to print a trace of its action (tokens shifted, rules reduced, etc). The trace mentions rule numbers and state numbers that can be interpreted by looking at the file grammar.output generated by ocamlyacc -v.

17.6 A complete example

The all-time favorite: a desk calculator. This program reads arithmetic expressions on standard input, one per line, and prints their values. Here is the grammar definition:

        /* File parser.mly */
        %token <int> INT
        %token PLUS MINUS TIMES DIV
        %token LPAREN RPAREN
        %token EOL
        %left PLUS MINUS        /* lowest precedence */
        %left TIMES DIV         /* medium precedence */
        %nonassoc UMINUS        /* highest precedence */
        %start main             /* the entry point */
        %type <int> main
        %%
        main:
            expr EOL                { $1 }
        ;
        expr:
            INT                     { $1 }
          | LPAREN expr RPAREN      { $2 }
          | expr PLUS expr          { $1 + $3 }
          | expr MINUS expr         { $1 - $3 }
          | expr TIMES expr         { $1 * $3 }
          | expr DIV expr           { $1 / $3 }
          | MINUS expr %prec UMINUS { - $2 }
        ;

Here is the definition for the corresponding lexer:

        (* File lexer.mll *)
        {
        open Parser        (* The type token is defined in parser.mli *)
        exception Eof
        }
        rule token = parse
            [' ' '\t']     { token lexbuf }     (* skip blanks *)
          | ['\n' ]        { EOL }
          | ['0'-'9']+ as lxm { INT(int_of_string lxm) }
          | '+'            { PLUS }
          | '-'            { MINUS }
          | '*'            { TIMES }
          | '/'            { DIV }
          | '('            { LPAREN }
          | ')'            { RPAREN }
          | eof            { raise Eof }

Here is the main program, that combines the parser with the lexer:

        (* File calc.ml *)
        let _ =
          try
            let lexbuf = Lexing.from_channel stdin in
            while true do
              let result = Parser.main Lexer.token lexbuf in
                print_int result; print_newline(); flush stdout
            done
          with Lexer.Eof ->
            exit 0

To compile everything, execute:

        ocamllex lexer.mll       # generates lexer.ml
        ocamlyacc parser.mly     # generates parser.ml and parser.mli
        ocamlc -c parser.mli
        ocamlc -c lexer.ml
        ocamlc -c parser.ml
        ocamlc -c calc.ml
        ocamlc -o calc lexer.cmo parser.cmo calc.cmo

17.7 Common errors

ocamllex: transition table overflow, automaton is too big

The deterministic automata generated by ocamllex are limited to at most 32767 transitions. The message above indicates that your lexer definition is too complex and overflows this limit. This is commonly caused by lexer definitions that have separate rules for each of the alphabetic keywords of the language, as in the following example.

rule token = parse
  "keyword1"   { KWD1 }
| "keyword2"   { KWD2 }
| ...
| "keyword100" { KWD100 }
| ['A'-'Z' 'a'-'z'] ['A'-'Z' 'a'-'z' '0'-'9' '_'] * as id
               { IDENT id}

To keep the generated automata small, rewrite those definitions with only one general “identifier” rule, followed by a hashtable lookup to separate keywords from identifiers:

{ let keyword_table = Hashtbl.create 53
  let _ =
    List.iter (fun (kwd, tok) -> Hashtbl.add keyword_table kwd tok)
              [ "keyword1", KWD1;
                "keyword2", KWD2; ...
                "keyword100", KWD100 ]
}
rule token = parse
  ['A'-'Z' 'a'-'z'] ['A'-'Z' 'a'-'z' '0'-'9' '_'] * as id
               { try
                   Hashtbl.find keyword_table id
                 with Not_found ->
                   IDENT id }
ocamllex: Position memory overflow, too many bindings
The deterministic automata generated by ocamllex maintain a table of positions inside the scanned lexer buffer. The size of this table is limited to at most 255 cells. This error should not show up in normal situations.
ocamlyacc: concurrency safety

Parsers generated by ocamlyacc are not thread-safe. Those parsers rely on an internal work state which is shared by all ocamlyacc generated parsers. The menhir parser generator is a better option if you want thread-safe parsers.

Chapter 18 Dependency generator (ocamldep)

The ocamldep command scans a set of OCaml source files (.ml and .mli files) for references to external compilation units, and outputs dependency lines in a format suitable for the make utility. This ensures that make will compile the source files in the correct order, and recompile those files that need to when a source file is modified.

The typical usage is:

        ocamldep options *.mli *.ml > .depend

where *.mli *.ml expands to all source files in the current directory and .depend is the file that should contain the dependencies. (See below for a typical Makefile.)

Dependencies are generated both for compiling with the bytecode compiler ocamlc and with the native-code compiler ocamlopt.

18.1 Options

The following command-line options are recognized by ocamldep.

-absname
Show absolute filenames in error messages.
-all
Generate dependencies on all required files, rather than assuming implicit dependencies.
-allow-approx
Allow falling back on a lexer-based approximation when parsing fails.
-args filename
Read additional newline-terminated command line arguments from filename.
-args0 filename
Read additional null character terminated command line arguments from filename.
-as-map
For the following files, do not include delayed dependencies for module aliases. This option assumes that they are compiled using options -no-alias-deps -w -49, and that those files or their interface are passed with the -map option when computing dependencies for other files. Note also that for dependencies to be correct in the implementation of a map file, its interface should not coerce any of the aliases it contains.
-debug-map
Dump the delayed dependency map for each map file.
-I directory
Add the given directory to the list of directories searched for source files. If a source file foo.ml mentions an external compilation unit Bar, a dependency on that unit’s interface bar.cmi is generated only if the source for bar is found in the current directory or in one of the directories specified with -I. Otherwise, Bar is assumed to be a module from the standard library, and no dependencies are generated. For programs that span multiple directories, it is recommended to pass ocamldep the same -I options that are passed to the compiler.
-nocwd
Do not add current working directory to the list of include directories.
-impl file
Process file as a .ml file.
-intf file
Process file as a .mli file.
-map file
Read and propagate the delayed dependencies for module aliases in file, so that the following files will depend on the exported aliased modules if they use them. See the example below.
-ml-synonym .ext
Consider the given extension (with leading dot) to be a synonym for .ml.
-mli-synonym .ext
Consider the given extension (with leading dot) to be a synonym for .mli.
-modules
Output raw dependencies of the form
      filename: Module1 Module2 ... ModuleN
where Module1, 
, ModuleN are the names of the compilation units referenced within the file filename, but these names are not resolved to source file names. Such raw dependencies cannot be used by make, but can be post-processed by other tools such as Omake.
-native
Generate dependencies for a pure native-code program (no bytecode version). When an implementation file (.ml file) has no explicit interface file (.mli file), ocamldep generates dependencies on the bytecode compiled file (.cmo file) to reflect interface changes. This can cause unnecessary bytecode recompilations for programs that are compiled to native-code only. The flag -native causes dependencies on native compiled files (.cmx) to be generated instead of on .cmo files. (This flag makes no difference if all source files have explicit .mli interface files.)
-one-line
Output one line per file, regardless of the length.
-open module
Assume that module module is opened before parsing each of the following files.
-pp command
Cause ocamldep to call the given command as a preprocessor for each source file.
-ppx command
Pipe abstract syntax trees through preprocessor command.
-shared
Generate dependencies for native plugin files (.cmxs) in addition to native compiled files (.cmx).
-slash
Under Windows, use a forward slash (/) as the path separator instead of the usual backward slash (\). Under Unix, this option does nothing.
-sort
Sort files according to their dependencies.
-version
Print version string and exit.
-vnum
Print short version number and exit.
-help or --help
Display a short usage summary and exit.

18.2 A typical Makefile

Here is a template Makefile for a OCaml program.

OCAMLC=ocamlc
OCAMLOPT=ocamlopt
OCAMLDEP=ocamldep
INCLUDES=                 # all relevant -I options here
OCAMLFLAGS=$(INCLUDES)    # add other options for ocamlc here
OCAMLOPTFLAGS=$(INCLUDES) # add other options for ocamlopt here

# prog1 should be compiled to bytecode, and is composed of three
# units: mod1, mod2 and mod3.

# The list of object files for prog1
PROG1_OBJS=mod1.cmo mod2.cmo mod3.cmo

prog1: $(PROG1_OBJS)
        $(OCAMLC) -o prog1 $(OCAMLFLAGS) $(PROG1_OBJS)

# prog2 should be compiled to native-code, and is composed of two
# units: mod4 and mod5.

# The list of object files for prog2
PROG2_OBJS=mod4.cmx mod5.cmx

prog2: $(PROG2_OBJS)
        $(OCAMLOPT) -o prog2 $(OCAMLFLAGS) $(PROG2_OBJS)

# Common rules

%.cmo: %.ml
        $(OCAMLC) $(OCAMLFLAGS) -c $<

%.cmi: %.mli
        $(OCAMLC) $(OCAMLFLAGS) -c $<

%.cmx: %.ml
        $(OCAMLOPT) $(OCAMLOPTFLAGS) -c $<

# Clean up
clean:
        rm -f prog1 prog2
        rm -f *.cm[iox]

# Dependencies
depend:
        $(OCAMLDEP) $(INCLUDES) *.mli *.ml > .depend

include .depend

If you use module aliases to give shorter names to modules, you need to change the above definitions. Assuming that your map file is called mylib.mli, here are minimal modifications.

OCAMLFLAGS=$(INCLUDES) -open Mylib

mylib.cmi: mylib.mli
        $(OCAMLC) $(INCLUDES) -no-alias-deps -w -49 -c $<

depend:
        $(OCAMLDEP) $(INCLUDES) -map mylib.mli $(PROG1_OBJS:.cmo=.ml) > .depend

Note that in this case you should not compute dependencies for mylib.mli together with the other files, hence the need to pass explicitly the list of files to process. If mylib.mli itself has dependencies, you should compute them using -as-map.

Chapter 19 The documentation generator (ocamldoc)

This chapter describes OCamldoc, a tool that generates documentation from special comments embedded in source files. The comments used by OCamldoc are of the form (**
*) and follow the format described in section 19.2.

OCamldoc can produce documentation in various formats: HTML, LATEX, TeXinfo, Unix man pages, and dot dependency graphs. Moreover, users can add their own custom generators, as explained in section 19.3.

In this chapter, we use the word element to refer to any of the following parts of an OCaml source file: a type declaration, a value, a module, an exception, a module type, a type constructor, a record field, a class, a class type, a class method, a class value or a class inheritance clause.

19.1 Usage

19.1.1 Invocation

OCamldoc is invoked via the command ocamldoc, as follows:

        ocamldoc options sourcefiles

Options for choosing the output format

The following options determine the format for the generated documentation.

-html
Generate documentation in HTML default format. The generated HTML pages are stored in the current directory, or in the directory specified with the -d option. You can customize the style of the generated pages by editing the generated style.css file, or by providing your own style sheet using option -css-style. The file style.css is not generated if it already exists or if -css-style is used.
-latex
Generate documentation in LATEX default format. The generated LATEX document is saved in file ocamldoc.out, or in the file specified with the -o option. The document uses the style file ocamldoc.sty. This file is generated when using the -latex option, if it does not already exist. You can change this file to customize the style of your LATEX documentation.
-texi
Generate documentation in TeXinfo default format. The generated LATEX document is saved in file ocamldoc.out, or in the file specified with the -o option.
-man
Generate documentation as a set of Unix man pages. The generated pages are stored in the current directory, or in the directory specified with the -d option.
-dot
Generate a dependency graph for the toplevel modules, in a format suitable for displaying and processing by dot. The dot tool is available from https://graphviz.org/. The textual representation of the graph is written to the file ocamldoc.out, or to the file specified with the -o option. Use dot ocamldoc.out to display it.
-g file.cm[o,a,xs]
Dynamically load the given file, which defines a custom documentation generator. See section 19.4.1. This option is supported by the ocamldoc command (to load .cmo and .cma files) and by its native-code version ocamldoc.opt (to load .cmxs files). If the given file is a simple one and does not exist in the current directory, then ocamldoc looks for it in the custom generators default directory, and in the directories specified with optional -i options.
-customdir
Display the custom generators default directory.
-i directory
Add the given directory to the path where to look for custom generators.

General options

-d dir
Generate files in directory dir, rather than the current directory.
-dump file
Dump collected information into file. This information can be read with the -load option in a subsequent invocation of ocamldoc.
-hide modules
Hide the given complete module names in the generated documentation. modules is a list of complete module names separated by ’,’, without blanks. For instance: Stdlib,M2.M3.
-inv-merge-ml-mli
Reverse the precedence of implementations and interfaces when merging. All elements in implementation files are kept, and the -m option indicates which parts of the comments in interface files are merged with the comments in implementation files.
-keep-code
Always keep the source code for values, methods and instance variables, when available.
-load file
Load information from file, which has been produced by ocamldoc -dump. Several -load options can be given.
-m flags
Specify merge options between interfaces and implementations. (see section 19.1.2 for details). flags can be one or several of the following characters:
d
merge description
a
merge @author
v
merge @version
l
merge @see
s
merge @since
b
merge @before
o
merge @deprecated
p
merge @param
e
merge @raise
r
merge @return
A
merge everything
-no-custom-tags
Do not allow custom @-tags (see section 19.2.5).
-no-stop
Keep elements placed after/between the (**/**) special comment(s) (see section 19.2).
-o file
Output the generated documentation to file instead of ocamldoc.out. This option is meaningful only in conjunction with the -latex, -texi, or -dot options.
-pp command
Pipe sources through preprocessor command.
-impl filename
Process the file filename as an implementation file, even if its extension is not .ml.
-intf filename
Process the file filename as an interface file, even if its extension is not .mli.
-text filename
Process the file filename as a text file, even if its extension is not .txt.
-sort
Sort the list of top-level modules before generating the documentation.
-stars
Remove blank characters until the first asterisk (’*’) in each line of comments.
-t title
Use title as the title for the generated documentation.
-intro file
Use content of file as ocamldoc text to use as introduction (HTML, LATEX and TeXinfo only). For HTML, the file is used to create the whole index.html file.
-v
Verbose mode. Display progress information.
-version
Print version string and exit.
-vnum
Print short version number and exit.
-warn-error
Treat Ocamldoc warnings as errors.
-hide-warnings
Do not print OCamldoc warnings.
-help or --help
Display a short usage summary and exit.

Type-checking options

OCamldoc calls the OCaml type-checker to obtain type information. The following options impact the type-checking phase. They have the same meaning as for the ocamlc and ocamlopt commands.

-I directory
Add directory to the list of directories search for compiled interface files (.cmi files).
-nolabels
Ignore non-optional labels in types.
-rectypes
Allow arbitrary recursive types. (See the -rectypes option to ocamlc.)

Options for generating HTML pages

The following options apply in conjunction with the -html option:

-all-params
Display the complete list of parameters for functions and methods.
-charset charset
Add information about character encoding being charset (default is iso-8859-1).
-colorize-code
Colorize the OCaml code enclosed in [ ] and {[ ]}, using colors to emphasize keywords, etc. If the code fragments are not syntactically correct, no color is added.
-css-style filename
Use filename as the Cascading Style Sheet file.
-index-only
Generate only index files.
-short-functors
Use a short form to display functors:
module M : functor (A:Module) -> functor (B:Module2) -> sig .. end
is displayed as:
module M (A:Module) (B:Module2) : sig .. end

Options for generating LATEX files

The following options apply in conjunction with the -latex option:

-latex-value-prefix prefix
Give a prefix to use for the labels of the values in the generated LATEX document. The default prefix is the empty string. You can also use the options -latex-type-prefix, -latex-exception-prefix, -latex-module-prefix, -latex-module-type-prefix, -latex-class-prefix, -latex-class-type-prefix, -latex-attribute-prefix and -latex-method-prefix.

These options are useful when you have, for example, a type and a value with the same name. If you do not specify prefixes, LATEX will complain about multiply defined labels.

-latextitle n,style
Associate style number n to the given LATEX sectioning command style, e.g. section or subsection. (LATEX only.) This is useful when including the generated document in another LATEX document, at a given sectioning level. The default association is 1 for section, 2 for subsection, 3 for subsubsection, 4 for paragraph and 5 for subparagraph.
-noheader
Suppress header in generated documentation.
-notoc
Do not generate a table of contents.
-notrailer
Suppress trailer in generated documentation.
-sepfiles
Generate one .tex file per toplevel module, instead of the global ocamldoc.out file.

Options for generating TeXinfo files

The following options apply in conjunction with the -texi option:

-esc8
Escape accented characters in Info files.
-info-entry
Specify Info directory entry.
-info-section
Specify section of Info directory.
-noheader
Suppress header in generated documentation.
-noindex
Do not build index for Info files.
-notrailer
Suppress trailer in generated documentation.

Options for generating dot graphs

The following options apply in conjunction with the -dot option:

-dot-colors colors
Specify the colors to use in the generated dot code. When generating module dependencies, ocamldoc uses different colors for modules, depending on the directories in which they reside. When generating types dependencies, ocamldoc uses different colors for types, depending on the modules in which they are defined. colors is a list of color names separated by ’,’, as in Red,Blue,Green. The available colors are the ones supported by the dot tool.
-dot-include-all
Include all modules in the dot output, not only modules given on the command line or loaded with the -load option.
-dot-reduce
Perform a transitive reduction of the dependency graph before outputting the dot code. This can be useful if there are a lot of transitive dependencies that clutter the graph.
-dot-types
Output dot code describing the type dependency graph instead of the module dependency graph.

Options for generating man files

The following options apply in conjunction with the -man option:

-man-mini
Generate man pages only for modules, module types, classes and class types, instead of pages for all elements.
-man-suffix suffix
Set the suffix used for generated man filenames. Default is ’3o’, as in List.3o.
-man-section section
Set the section number used for generated man filenames. Default is ’3’.

19.1.2 Merging of module information

Information on a module can be extracted either from the .mli or .ml file, or both, depending on the files given on the command line. When both .mli and .ml files are given for the same module, information extracted from these files is merged according to the following rules:

19.1.3 Coding rules

The following rules must be respected in order to avoid name clashes resulting in cross-reference errors:

19.2 Syntax of documentation comments

Comments containing documentation material are called special comments and are written between (** and *). Special comments must start exactly with (**. Comments beginning with ( and more than two * are ignored.

19.2.1 Placement of documentation comments

OCamldoc can associate comments to some elements of the language encountered in the source files. The association is made according to the locations of comments with respect to the language elements. The locations of comments in .mli and .ml files are different.

Comments in .mli files

A special comment is associated to an element if it is placed before or after the element.
A special comment before an element is associated to this element if :

A special comment after an element is associated to this element if there is no blank line or comment between the special comment and the element.

There are two exceptions: for constructors and record fields in type definitions, the associated comment can only be placed after the constructor or field definition, without blank lines or other comments between them. The special comment for a constructor with another constructor following must be placed before the ’|’ character separating the two constructors.

The following sample interface file foo.mli illustrates the placement rules for comments in .mli files.

(** The first special comment of the file is the comment associated with the whole module.*) (** Special comments can be placed between elements and are kept by the OCamldoc tool, but are not associated to any element. @-tags in these comments are ignored.*) (*******************************************************************) (** Comments like the one above, with more than two asterisks, are ignored. *) (** The comment for function f. *) val f : int -> int -> int (** The continuation of the comment for function f. *) (** Comment for exception My_exception, even with a simple comment between the special comment and the exception.*) (* Hello, I'm a simple comment :-) *) exception My_exception of (int -> int) * int (** Comment for type weather *) type weather = | Rain of int (** The comment for constructor Rain *) | Sun (** The comment for constructor Sun *) (** Comment for type weather2 *) type weather2 = | Rain of int (** The comment for constructor Rain *) | Sun (** The comment for constructor Sun *) (** I can continue the comment for type weather2 here because there is already a comment associated to the last constructor.*) (** The comment for type my_record *) type my_record = { foo : int ; (** Comment for field foo *) bar : string ; (** Comment for field bar *) } (** Continuation of comment for type my_record *) (** Comment for foo *) val foo : string (** This comment is associated to foo and not to bar. *) val bar : string (** This comment is associated to bar. *) (** The comment for class my_class *) class my_class : object (** A comment to describe inheritance from cl *) inherit cl (** The comment for attribute tutu *) val mutable tutu : string (** The comment for attribute toto. *) val toto : int (** This comment is not attached to titi since there is a blank line before titi, but is kept as a comment in the class. *) val titi : string (** Comment for method toto *) method toto : string (** Comment for method m *) method m : float -> int end (** The comment for the class type my_class_type *) class type my_class_type = object (** The comment for variable x. *) val mutable x : int (** The comment for method m. *) method m : int -> int end (** The comment for module Foo *) module Foo : sig (** The comment for x *) val x : int (** A special comment that is kept but not associated to any element *) end (** The comment for module type my_module_type. *) module type my_module_type = sig (** The comment for value x. *) val x : int (** The comment for module M. *) module M : sig (** The comment for value y. *) val y : int (* ... *) end end

Comments in .ml files

A special comment is associated to an element if it is placed before the element and there is no blank line between the comment and the element. Meanwhile, there can be a simple comment between the special comment and the element. There are two exceptions, for constructors and record fields in type definitions, whose associated comment must be placed after the constructor or field definition, without blank line between them. The special comment for a constructor with another constructor following must be placed before the ’|’ character separating the two constructors.

The following example of file toto.ml shows where to place comments in a .ml file.

(** The first special comment of the file is the comment associated to the whole module. *) (** The comment for function f *) let f x y = x + y (** This comment is not attached to any element since there is another special comment just before the next element. *) (** Comment for exception My_exception, even with a simple comment between the special comment and the exception.*) (* A simple comment. *) exception My_exception of (int -> int) * int (** Comment for type weather *) type weather = | Rain of int (** The comment for constructor Rain *) | Sun (** The comment for constructor Sun *) (** The comment for type my_record *) type my_record = { foo : int ; (** Comment for field foo *) bar : string ; (** Comment for field bar *) } (** The comment for class my_class *) class my_class = object (** A comment to describe inheritance from cl *) inherit cl (** The comment for the instance variable tutu *) val mutable tutu = "tutu" (** The comment for toto *) val toto = 1 val titi = "titi" (** Comment for method toto *) method toto = tutu ^ "!" (** Comment for method m *) method m (f : float) = 1 end (** The comment for class type my_class_type *) class type my_class_type = object (** The comment for the instance variable x. *) val mutable x : int (** The comment for method m. *) method m : int -> int end (** The comment for module Foo *) module Foo = struct (** The comment for x *) let x = 0 (** A special comment in the class, but not associated to any element. *) end (** The comment for module type my_module_type. *) module type my_module_type = sig (* Comment for value x. *) val x : int (* ... *) end

19.2.2 The Stop special comment

The special comment (**/**) tells OCamldoc to discard elements placed after this comment, up to the end of the current class, class type, module or module type, or up to the next stop comment. For instance:

class type foo = object (** comment for method m *) method m : string (**/**) (** This method won't appear in the documentation *) method bar : int end (** This value appears in the documentation, since the Stop special comment in the class does not affect the parent module of the class.*) val foo : string (**/**) (** The value bar does not appear in the documentation.*) val bar : string (**/**) (** The type t appears since in the documentation since the previous stop comment toggled off the "no documentation mode". *) type t = string

The -no-stop option to ocamldoc causes the Stop special comments to be ignored.

19.2.3 Syntax of documentation comments

The inside of documentation comments (**
*) consists of free-form text with optional formatting annotations, followed by optional tags giving more specific information about parameters, version, authors, 
 The tags are distinguished by a leading @ character. Thus, a documentation comment has the following shape:

(** The comment begins with a description, which is text formatted
   according to the rules described in the next section.
   The description continues until the first non-escaped '@' character.
   @author Mr Smith
   @param x description for parameter x
*)

Some elements support only a subset of all @-tags. Tags that are not relevant to the documented element are simply ignored. For instance, all tags are ignored when documenting type constructors, record fields, and class inheritance clauses. Similarly, a @param tag on a class instance variable is ignored.

At last, (**) is the empty documentation comment.

19.2.4 Text formatting

Here is the BNF grammar for the simple markup language used to format text descriptions.

text::= {text-element}+
 
inline-text::= {inline-text-element}+
 
text-element::=
∣inline-text-element 
∣blank-lineforce a new line.


inline-text-element::=
∣ { { 0 
 9 }+ inline-text } format text as a section header; the integer following { indicates the sectioning level.
∣ { { 0 
 9 }+ : label inline-text } same, but also associate the name label to the current point. This point can be referenced by its fully-qualified label in a {! command, just like any other element.
∣ {b inline-text } set text in bold.
∣ {i inline-text } set text in italic.
∣ {e inline-text } emphasize text.
∣ {C inline-text } center text.
∣ {L inline-text } left align text.
∣ {R inline-text } right align text.
∣ {ul list } build a list.
∣ {ol list } build an enumerated list.
∣ {{: string } inline-text } put a link to the given address (given as string) on the given text.
∣ [ string ] set the given string in source code style.
∣ {[ string ]} set the given string in preformatted source code style.
∣ {v string v} set the given string in verbatim style.
∣ {% string %} target-specific content (LATEX code by default, see details in 19.2.4.4)
∣ {! string } insert a cross-reference to an element (see section 19.2.4.2 for the syntax of cross-references).
∣ {{! string } inline-text } insert a cross-reference with the given text.
∣ {!modules: string string ... } insert an index table for the given module names. Used in HTML only.
∣ {!indexlist} insert a table of links to the various indexes (types, values, modules, ...). Used in HTML only.
∣ {^ inline-text } set text in superscript.
∣ {_ inline-text } set text in subscript.
∣escaped-stringtypeset the given string as is; special characters (’{’, ’}’, ’[’, ’]’ and ’@’) must be escaped by a ’\’


19.2.4.1 List formatting

list::= 
 âˆŁâ€„{ {- inline-text } }+
 âˆŁâ€„{ {li inline-text } }+

A shortcut syntax exists for lists and enumerated lists:

(** Here is a {b list}
- item 1
- item 2
- item 3

The list is ended by the blank line.*)

is equivalent to:

(** Here is a {b list}
{ul {- item 1}
{- item 2}
{- item 3}}
The list is ended by the blank line.*)

The same shortcut is available for enumerated lists, using ’+’ instead of ’-’. Note that only one list can be defined by this shortcut in nested lists.

19.2.4.2 Cross-reference formatting

Cross-references are fully qualified element names, as in the example {!Foo.Bar.t}. This is an ambiguous reference as it may designate a type name, a value name, a class name, etc. It is possible to make explicit the intended syntactic class, using {!type:Foo.Bar.t} to designate a type, and {!val:Foo.Bar.t} a value of the same name.

The list of possible syntactic class is as follows:

tagsyntactic class
module:module
modtype:module type
class:class
classtype:class type
val:value
type:type
exception:exception
attribute:attribute
method:class method
section:ocamldoc section
const:variant constructor
recfield:record field

In the case of variant constructors or record fields, the constructor or field name should be preceded by the name of the corresponding type to avoid the ambiguity of several types having the same constructor names. For example, the constructor Node of the type tree will be referenced as {!tree.Node} or {!const:tree.Node}, or possibly {!Mod1.Mod2.tree.Node} from outside the module.

19.2.4.3 First sentence

In the description of a value, type, exception, module, module type, class or class type, the first sentence is sometimes used in indexes, or when just a part of the description is needed. The first sentence is composed of the first characters of the description, until

outside of the following text formatting : {ul list } , {ol list } , [ string ] , {[ string ]} , {v string v} , {% string %} , {! string } , {^ text } , {_ text } .

19.2.4.4 Target-specific formatting

The content inside {%foo: ... %} is target-specific and will be interpreted only by the backend foo, and ignored by other backends. The backends of the distribution are latex, html, texi and man. If no target is specified (syntax {% ... %}), latex is chosen by default. Custom generators may support their own target prefix.

19.2.4.5 Recognized HTML tags

The HTML tags <b>..</b>, <code>..</code>, <i>..</i>, <ul>..</ul>, <ol>..</ol>, <li>..</li>, <center>..</center> and <h[0-9]>..</h[0-9]> can be used instead of, respectively, {b ..} , [..] , {i ..} , {ul ..} , {ol ..} , {li ..} , {C ..} and {[0-9] ..}.

19.2.5 Documentation tags (@-tags)

Predefined tags

The following table gives the list of predefined @-tags, with their syntax and meaning.

@author string The author of the element. One author per @author tag. There may be several @author tags for the same element.
@deprecated text The text should describe when the element was deprecated, what to use as a replacement, and possibly the reason for deprecation.
@param id text Associate the given description (text) to the given parameter name id. This tag is used for functions, methods, classes and functors.
@raise Exc text Explain that the element may raise the exception Exc.
@return text Describe the return value and its possible values. This tag is used for functions and methods.
@see < URL > text Add a reference to the URL with the given text as comment.
@see 'filename' text Add a reference to the given file name (written between single quotes), with the given text as comment.
@see "document-name" text Add a reference to the given document name (written between double quotes), with the given text as comment.
@since string Indicate when the element was introduced.
@before version text Associate the given description (text) to the given version in order to document compatibility issues.
@version string The version number for the element.

Custom tags

You can use custom tags in the documentation comments, but they will have no effect if the generator used does not handle them. To use a custom tag, for example foo, just put @foo with some text in your comment, as in:

(** My comment to show you a custom tag.
@foo this is the text argument to the [foo] custom tag.
*)

To handle custom tags, you need to define a custom generator, as explained in section 19.3.2.

19.3 Custom generators

OCamldoc operates in two steps:

  1. analysis of the source files;
  2. generation of documentation, through a documentation generator, which is an object of class Odoc_args.class_generator.

Users can provide their own documentation generator to be used during step 2 instead of the default generators. All the information retrieved during the analysis step is available through the Odoc_info module, which gives access to all the types and functions representing the elements found in the given modules, with their associated description.

The files you can use to define custom generators are installed in the ocamldoc sub-directory of the OCaml standard library.

19.3.1 The generator modules

The type of a generator module depends on the kind of generated documentation. Here is the list of generator module types, with the name of the generator class in the module :

That is, to define a new generator, one must implement a module with the expected signature, and with the given generator class, providing the generate method as entry point to make the generator generates documentation for a given list of modules :

        method generate : Odoc_info.Module.t_module list -> unit

This method will be called with the list of analysed and possibly merged Odoc_info.t_module structures.

It is recommended to inherit from the current generator of the same kind as the one you want to define. Doing so, it is possible to load various custom generators to combine improvements brought by each one.

This is done using first class modules (see chapter 12.5).

The easiest way to define a custom generator is the following this example, here extending the current HTML generator. We don’t have to know if this is the original HTML generator defined in ocamldoc or if it has been extended already by a previously loaded custom generator :

module Generator (G : Odoc_html.Html_generator) =
struct
  class html =
    object(self)
      inherit G.html as html
      (* ... *)

      method generate module_list =
        (* ... *)
        ()

      (* ... *)
  end
end;;

let _ = Odoc_args.extend_html_generator (module Generator : Odoc_gen.Html_functor);;

To know which methods to override and/or which methods are available, have a look at the different base implementations, depending on the kind of generator you are extending :

19.3.2 Handling custom tags

Making a custom generator handle custom tags (see 19.2.5) is very simple.

For HTML

Here is how to develop a HTML generator handling your custom tags.

The class Odoc_html.Generator.html inherits from the class Odoc_html.info, containing a field tag_functions which is a list pairs composed of a custom tag (e.g. "foo") and a function taking a text and returning HTML code (of type string). To handle a new tag bar, extend the current HTML generator and complete the tag_functions field:

module Generator (G : Odoc_html.Html_generator) =
struct
  class html =
    object(self)
      inherit G.html

      (** Return HTML code for the given text of a bar tag. *)
      method html_of_bar t = (* your code here *)

      initializer
        tag_functions <- ("bar", self#html_of_bar) :: tag_functions
  end
end
let _ = Odoc_args.extend_html_generator (module Generator : Odoc_gen.Html_functor);;

Another method of the class Odoc_html.info will look for the function associated to a custom tag and apply it to the text given to the tag. If no function is associated to a custom tag, then the method prints a warning message on stderr.

For other generators

You can act the same way for other kinds of generators.

19.4 Adding command line options

The command line analysis is performed after loading the module containing the documentation generator, thus allowing command line options to be added to the list of existing ones. Adding an option can be done with the function

        Odoc_args.add_option : string * Arg.spec * string -> unit

Note: Existing command line options can be redefined using this function.

19.4.1 Compilation and usage

Defining a custom generator class in one file

Let custom.ml be the file defining a new generator class. Compilation of custom.ml can be performed by the following command :

        ocamlc -I +ocamldoc -c custom.ml

The file custom.cmo is created and can be used this way :

        ocamldoc -g custom.cmo other-options source-files

Options selecting a built-in generator to ocamldoc, such as -html, have no effect if a custom generator of the same kind is provided using -g. If the kinds do not match, the selected built-in generator is used and the custom one is ignored.

Defining a custom generator class in several files

It is possible to define a generator class in several modules, which are defined in several files file1.ml[i], file2.ml[i], ..., filen.ml[i]. A .cma library file must be created, including all these files.

The following commands create the custom.cma file from files file1.ml[i], ..., filen.ml[i] :

ocamlc -I +ocamldoc -c file1.ml[i]
ocamlc -I +ocamldoc -c file2.ml[i]
...
ocamlc -I +ocamldoc -c filen.ml[i]
ocamlc -o custom.cma -a file1.cmo file2.cmo ... filen.cmo

Then, the following command uses custom.cma as custom generator:

        ocamldoc -g custom.cma other-options source-files

Chapter 20 The debugger (ocamldebug)

This chapter describes the OCaml source-level replay debugger ocamldebug.

Unix:  The debugger is available on Unix systems that provide BSD sockets.
Windows:  The debugger is available under the Cygwin port of OCaml, but not under the native Win32 ports.

20.1 Compiling for debugging

Before the debugger can be used, the program must be compiled and linked with the -g option: all .cmo and .cma files that are part of the program should have been created with ocamlc -g, and they must be linked together with ocamlc -g.

Compiling with -g entails no penalty on the running time of programs: object files and bytecode executable files are bigger and take longer to produce, but the executable files run at exactly the same speed as if they had been compiled without -g.

20.2 Invocation

20.2.1 Starting the debugger

The OCaml debugger is invoked by running the program ocamldebug with the name of the bytecode executable file as first argument:

        ocamldebug [options] program [arguments]

The arguments following program are optional, and are passed as command-line arguments to the program being debugged. (See also the set arguments command.)

The following command-line options are recognized:

-c count
Set the maximum number of simultaneously live checkpoints to count.
-cd dir
Run the debugger program from the working directory dir, instead of the current directory. (See also the cd command.)
-emacs
Tell the debugger it is executed under Emacs. (See section 20.10 for information on how to run the debugger under Emacs.)
-I directory
Add directory to the list of directories searched for source files and compiled files. (See also the directory command.)
-s socket
Use socket for communicating with the debugged program. See the description of the command set socket (section 20.8.8) for the format of socket.
-version
Print version string and exit.
-vnum
Print short version number and exit.
-help or --help
Display a short usage summary and exit.

20.2.2 Initialization file

On start-up, the debugger will read commands from an initialization file before giving control to the user. The default file is .ocamldebug in the current directory if it exists, otherwise .ocamldebug in the user’s home directory.

20.2.3 Exiting the debugger

The command quit exits the debugger. You can also exit the debugger by typing an end-of-file character (usually ctrl-D).

Typing an interrupt character (usually ctrl-C) will not exit the debugger, but will terminate the action of any debugger command that is in progress and return to the debugger command level.

20.3 Commands

A debugger command is a single line of input. It starts with a command name, which is followed by arguments depending on this name. Examples:

        run
        goto 1000
        set arguments arg1 arg2

A command name can be truncated as long as there is no ambiguity. For instance, go 1000 is understood as goto 1000, since there are no other commands whose name starts with go. For the most frequently used commands, ambiguous abbreviations are allowed. For instance, r stands for run even though there are others commands starting with r. You can test the validity of an abbreviation using the help command.

If the previous command has been successful, a blank line (typing just RET) will repeat it.

20.3.1 Getting help

The OCaml debugger has a simple on-line help system, which gives a brief description of each command and variable.

help
Print the list of commands.
help command
Give help about the command command.
help set variable, help show variable
Give help about the variable variable. The list of all debugger variables can be obtained with help set.
help info topic
Give help about topic. Use help info to get a list of known topics.

20.3.2 Accessing the debugger state

set variable value
Set the debugger variable variable to the value value.
show variable
Print the value of the debugger variable variable.
info subject
Give information about the given subject. For instance, info breakpoints will print the list of all breakpoints.

20.4 Executing a program

20.4.1 Events

Events are “interesting” locations in the source code, corresponding to the beginning or end of evaluation of “interesting” sub-expressions. Events are the unit of single-stepping (stepping goes to the next or previous event encountered in the program execution). Also, breakpoints can only be set at events. Thus, events play the role of line numbers in debuggers for conventional languages.

During program execution, a counter is incremented at each event encountered. The value of this counter is referred as the current time. Thanks to reverse execution, it is possible to jump back and forth to any time of the execution.

Here is where the debugger events (written ⋈) are located in the source code:

Exceptions: A function application followed by a function return is replaced by the compiler by a jump (tail-call optimization). In this case, no event is put after the function application.

20.4.2 Starting the debugged program

The debugger starts executing the debugged program only when needed. This allows setting breakpoints or assigning debugger variables before execution starts. There are several ways to start execution:

run
Run the program until a breakpoint is hit, or the program terminates.
goto 0
Load the program and stop on the first event.
goto time
Load the program and execute it until the given time. Useful when you already know approximately at what time the problem appears. Also useful to set breakpoints on function values that have not been computed at time 0 (see section 20.5).

The execution of a program is affected by certain information it receives when the debugger starts it, such as the command-line arguments to the program and its working directory. The debugger provides commands to specify this information (set arguments and cd). These commands must be used before program execution starts. If you try to change the arguments or the working directory after starting your program, the debugger will kill the program (after asking for confirmation).

20.4.3 Running the program

The following commands execute the program forward or backward, starting at the current time. The execution will stop either when specified by the command or when a breakpoint is encountered.

run
Execute the program forward from current time. Stops at next breakpoint or when the program terminates.
reverse
Execute the program backward from current time. Mostly useful to go to the last breakpoint encountered before the current time.
step [count]
Run the program and stop at the next event. With an argument, do it count times. If count is 0, run until the program terminates or a breakpoint is hit.
backstep [count]
Run the program backward and stop at the previous event. With an argument, do it count times.
next [count]
Run the program and stop at the next event, skipping over function calls. With an argument, do it count times.
previous [count]
Run the program backward and stop at the previous event, skipping over function calls. With an argument, do it count times.
finish
Run the program until the current function returns.
start
Run the program backward and stop at the first event before the current function invocation.

20.4.4 Time travel

You can jump directly to a given time, without stopping on breakpoints, using the goto command.

As you move through the program, the debugger maintains an history of the successive times you stop at. The last command can be used to revisit these times: each last command moves one step back through the history. That is useful mainly to undo commands such as step and next.

goto time
Jump to the given time.
last [count]
Go back to the latest time recorded in the execution history. With an argument, do it count times.
set history size
Set the size of the execution history.

20.4.5 Killing the program

kill
Kill the program being executed. This command is mainly useful if you wish to recompile the program without leaving the debugger.

20.5 Breakpoints

A breakpoint causes the program to stop whenever a certain point in the program is reached. It can be set in several ways using the break command. Breakpoints are assigned numbers when set, for further reference. The most comfortable way to set breakpoints is through the Emacs interface (see section 20.10).

break
Set a breakpoint at the current position in the program execution. The current position must be on an event (i.e., neither at the beginning, nor at the end of the program).
break function
Set a breakpoint at the beginning of function. This works only when the functional value of the identifier function has been computed and assigned to the identifier. Hence this command cannot be used at the very beginning of the program execution, when all identifiers are still undefined; use goto time to advance execution until the functional value is available.
break @ [module] line
Set a breakpoint in module module (or in the current module if module is not given), at the first event of line line.
break @ [module] line column
Set a breakpoint in module module (or in the current module if module is not given), at the event closest to line line, column column.
break @ [module] # character
Set a breakpoint in module module at the event closest to character number character.
break frag:pc, break pc
Set a breakpoint at code address frag:pc. The integer frag is the identifier of a code fragment, a set of modules that have been loaded at once, either initially or with the Dynlink module. The integer pc is the instruction counter within this code fragment. If frag is omitted, it defaults to 0, which is the code fragment of the program loaded initially.
delete [breakpoint-numbers]
Delete the specified breakpoints. Without argument, all breakpoints are deleted (after asking for confirmation).
info breakpoints
Print the list of all breakpoints.

20.6 The call stack

Each time the program performs a function application, it saves the location of the application (the return address) in a block of data called a stack frame. The frame also contains the local variables of the caller function. All the frames are allocated in a region of memory called the call stack. The command backtrace (or bt) displays parts of the call stack.

At any time, one of the stack frames is “selected” by the debugger; several debugger commands refer implicitly to the selected frame. In particular, whenever you ask the debugger for the value of a local variable, the value is found in the selected frame. The commands frame, up and down select whichever frame you are interested in.

When the program stops, the debugger automatically selects the currently executing frame and describes it briefly as the frame command does.

frame
Describe the currently selected stack frame.
frame frame-number
Select a stack frame by number and describe it. The frame currently executing when the program stopped has number 0; its caller has number 1; and so on up the call stack.
backtrace [count], bt [count]
Print the call stack. This is useful to see which sequence of function calls led to the currently executing frame. With a positive argument, print only the innermost count frames. With a negative argument, print only the outermost -count frames.
up [count]
Select and display the stack frame just “above” the selected frame, that is, the frame that called the selected frame. An argument says how many frames to go up.
down [count]
Select and display the stack frame just “below” the selected frame, that is, the frame that was called by the selected frame. An argument says how many frames to go down.

20.7 Examining variable values

The debugger can print the current value of simple expressions. The expressions can involve program variables: all the identifiers that are in scope at the selected program point can be accessed.

Expressions that can be printed are a subset of OCaml expressions, as described by the following grammar:

simple-expr::= lowercase-ident
 âˆŁâ€„{ capitalized-ident . } lowercase-ident
 âˆŁâ€„*
 âˆŁâ€„$ integer
 âˆŁâ€„simple-expr . lowercase-ident
 âˆŁâ€„simple-expr .( integer )
 âˆŁâ€„simple-expr .[ integer ]
 âˆŁâ€„! simple-expr
 âˆŁâ€„( simple-expr )

The first two cases refer to a value identifier, either unqualified or qualified by the path to the structure that define it. * refers to the result just computed (typically, the value of a function application), and is valid only if the selected event is an “after” event (typically, a function application). $ integer refer to a previously printed value. The remaining four forms select part of an expression: respectively, a record field, an array element, a string element, and the current contents of a reference.

print variables
Print the values of the given variables. print can be abbreviated as p.
display variables
Same as print, but limit the depth of printing to 1. Useful to browse large data structures without printing them in full. display can be abbreviated as d.

When printing a complex expression, a name of the form $integer is automatically assigned to its value. Such names are also assigned to parts of the value that cannot be printed because the maximal printing depth is exceeded. Named values can be printed later on with the commands p $integer or d $integer. Named values are valid only as long as the program is stopped. They are forgotten as soon as the program resumes execution.

set print_depth d
Limit the printing of values to a maximal depth of d.
set print_length l
Limit the printing of values to at most l nodes printed.

20.8 Controlling the debugger

20.8.1 Setting the program name and arguments

set program file
Set the program name to file.
set arguments arguments
Give arguments as command-line arguments for the program.

A shell is used to pass the arguments to the debugged program. You can therefore use wildcards, shell variables, and file redirections inside the arguments. To debug programs that read from standard input, it is recommended to redirect their input from a file (using set arguments < input-file), otherwise input to the program and input to the debugger are not properly separated, and inputs are not properly replayed when running the program backwards.

20.8.2 How programs are loaded

The loadingmode variable controls how the program is executed.

set loadingmode direct
The program is run directly by the debugger. This is the default mode.
set loadingmode runtime
The debugger execute the OCaml runtime ocamlrun on the program. Rarely useful; moreover it prevents the debugging of programs compiled in “custom runtime” mode.
set loadingmode manual
The user starts manually the program, when asked by the debugger. Allows remote debugging (see section 20.8.8).

20.8.3 Search path for files

The debugger searches for source files and compiled interface files in a list of directories, the search path. The search path initially contains the current directory . and the standard library directory. The directory command adds directories to the path.

Whenever the search path is modified, the debugger will clear any information it may have cached about the files.

directory directorynames
Add the given directories to the search path. These directories are added at the front, and will therefore be searched first.
directory directorynames for modulename
Same as directory directorynames, but the given directories will be searched only when looking for the source file of a module that has been packed into modulename.
directory
Reset the search path. This requires confirmation.

20.8.4 Working directory

Each time a program is started in the debugger, it inherits its working directory from the current working directory of the debugger. This working directory is initially whatever it inherited from its parent process (typically the shell), but you can specify a new working directory in the debugger with the cd command or the -cd command-line option.

cd directory
Set the working directory for ocamldebug to directory.
pwd
Print the working directory for ocamldebug.

20.8.5 Turning reverse execution on and off

In some cases, you may want to turn reverse execution off. This speeds up the program execution, and is also sometimes useful for interactive programs.

Normally, the debugger takes checkpoints of the program state from time to time. That is, it makes a copy of the current state of the program (using the Unix system call fork). If the variable checkpoints is set to off, the debugger will not take any checkpoints.

set checkpoints on/off
Select whether the debugger makes checkpoints or not.

20.8.6 Behavior of the debugger with respect to fork

When the program issues a call to fork, the debugger can either follow the child or the parent. By default, the debugger follows the parent process. The variable follow_fork_mode controls this behavior:

set follow_fork_mode child/parent
Select whether to follow the child or the parent in case of a call to fork.

20.8.7 Stopping execution when new code is loaded

The debugger is compatible with the Dynlink module. However, when an external module is not yet loaded, it is impossible to set a breakpoint in its code. In order to facilitate setting breakpoints in dynamically loaded code, the debugger stops the program each time new modules are loaded. This behavior can be disabled using the break_on_load variable:

set break_on_load on/off
Select whether to stop after loading new code.

20.8.8 Communication between the debugger and the program

The debugger communicate with the program being debugged through a Unix socket. You may need to change the socket name, for example if you need to run the debugger on a machine and your program on another.

set socket socket
Use socket for communication with the program. socket can be either a file name, or an Internet port specification host:port, where host is a host name or an Internet address in dot notation, and port is a port number on the host.

On the debugged program side, the socket name is passed through the CAML_DEBUG_SOCKET environment variable.

20.8.9 Fine-tuning the debugger

Several variables enables to fine-tune the debugger. Reasonable defaults are provided, and you should normally not have to change them.

set processcount count
Set the maximum number of checkpoints to count. More checkpoints facilitate going far back in time, but use more memory and create more Unix processes.

As checkpointing is quite expensive, it must not be done too often. On the other hand, backward execution is faster when checkpoints are taken more often. In particular, backward single-stepping is more responsive when many checkpoints have been taken just before the current time. To fine-tune the checkpointing strategy, the debugger does not take checkpoints at the same frequency for long displacements (e.g. run) and small ones (e.g. step). The two variables bigstep and smallstep contain the number of events between two checkpoints in each case.

set bigstep count
Set the number of events between two checkpoints for long displacements.
set smallstep count
Set the number of events between two checkpoints for small displacements.

The following commands display information on checkpoints and events:

info checkpoints
Print a list of checkpoints.
info events [module]
Print the list of events in the given module (the current module, by default).

20.8.10 User-defined printers

Just as in the toplevel system (section 14.2), the user can register functions for printing values of certain types. For technical reasons, the debugger cannot call printing functions that reside in the program being debugged. The code for the printing functions must therefore be loaded explicitly in the debugger.

load_printer "file-name"
Load in the debugger the indicated .cmo or .cma object file. The file is loaded in an environment consisting only of the OCaml standard library plus the definitions provided by object files previously loaded using load_printer. If this file depends on other object files not yet loaded, the debugger automatically loads them if it is able to find them in the search path. The loaded file does not have direct access to the modules of the program being debugged.
install_printer printer-name
Register the function named printer-name (a value path) as a printer for objects whose types match the argument type of the function. That is, the debugger will call printer-name when it has such an object to print. The printing function printer-name must use the Format library module to produce its output, otherwise its output will not be correctly located in the values printed by the toplevel loop.

The value path printer-name must refer to one of the functions defined by the object files loaded using load_printer. It cannot reference the functions of the program being debugged.

remove_printer printer-name
Remove the named function from the table of value printers.

20.9 Miscellaneous commands

list [module] [beginning] [end]
List the source of module module, from line number beginning to line number end. By default, 20 lines of the current module are displayed, starting 10 lines before the current position.
source filename
Read debugger commands from the script filename.

20.10 Running the debugger under Emacs

The most user-friendly way to use the debugger is to run it under Emacs with the OCaml mode available through MELPA and also at https://github.com/ocaml/caml-mode.

The OCaml debugger is started under Emacs by the command M-x camldebug, with argument the name of the executable file progname to debug. Communication with the debugger takes place in an Emacs buffer named *camldebug-progname*. The editing and history facilities of Shell mode are available for interacting with the debugger.

In addition, Emacs displays the source files containing the current event (the current position in the program execution) and highlights the location of the event. This display is updated synchronously with the debugger action.

The following bindings for the most common debugger commands are available in the *camldebug-progname* buffer:

C-c C-s
(command step): execute the program one step forward.
C-c C-k
(command backstep): execute the program one step backward.
C-c C-n
(command next): execute the program one step forward, skipping over function calls.
Middle mouse button
(command display): display named value. $n under mouse cursor (support incremental browsing of large data structures).
C-c C-p
(command print): print value of identifier at point.
C-c C-d
(command display): display value of identifier at point.
C-c C-r
(command run): execute the program forward to next breakpoint.
C-c C-v
(command reverse): execute the program backward to latest breakpoint.
C-c C-l
(command last): go back one step in the command history.
C-c C-t
(command backtrace): display backtrace of function calls.
C-c C-f
(command finish): run forward till the current function returns.
C-c <
(command up): select the stack frame below the current frame.
C-c >
(command down): select the stack frame above the current frame.

In all buffers in OCaml editing mode, the following debugger commands are also available:

C-x C-a C-b
(command break): set a breakpoint at event closest to point
C-x C-a C-p
(command print): print value of identifier at point
C-x C-a C-d
(command display): display value of identifier at point

Chapter 21 Profiling (ocamlprof)

This chapter describes how the execution of OCaml programs can be profiled, by recording how many times functions are called, branches of conditionals are taken, 


21.1 Compiling for profiling

Before profiling an execution, the program must be compiled in profiling mode, using the ocamlcp front-end to the ocamlc compiler (see chapter 13) or the ocamloptp front-end to the ocamlopt compiler (see chapter 16). When compiling modules separately, ocamlcp or ocamloptp must be used when compiling the modules (production of .cmo or .cmx files), and can also be used (though this is not strictly necessary) when linking them together.

Note

If a module (.ml file) doesn’t have a corresponding interface (.mli file), then compiling it with ocamlcp will produce object files (.cmi and .cmo) that are not compatible with the ones produced by ocamlc, which may lead to problems (if the .cmi or .cmo is still around) when switching between profiling and non-profiling compilations. To avoid this problem, you should always have a .mli file for each .ml file. The same problem exists with ocamloptp.

Note

To make sure your programs can be compiled in profiling mode, avoid using any identifier that begins with __ocaml_prof.

The amount of profiling information can be controlled through the -P option to ocamlcp or ocamloptp, followed by one or several letters indicating which parts of the program should be profiled:

a
all options
f
function calls : a count point is set at the beginning of each function body
i
if 
then 
else 
 : count points are set in both then branch and else branch
l
while, for loops: a count point is set at the beginning of the loop body
m
match branches: a count point is set at the beginning of the body of each branch
t
try 
with 
 branches: a count point is set at the beginning of the body of each branch

For instance, compiling with ocamlcp -P film profiles function calls, if
then
else
, loops and pattern matching.

Calling ocamlcp or ocamloptp without the -P option defaults to -P fm, meaning that only function calls and pattern matching are profiled.

Note

For compatibility with previous releases, ocamlcp also accepts the -p option, with the same arguments and behaviour as -P.

The ocamlcp and ocamloptp commands also accept all the options of the corresponding ocamlc or ocamlopt compiler, except the -pp (preprocessing) option.

21.2 Profiling an execution

Running an executable that has been compiled with ocamlcp or ocamloptp records the execution counts for the specified parts of the program and saves them in a file called ocamlprof.dump in the current directory.

If the environment variable OCAMLPROF_DUMP is set when the program exits, its value is used as the file name instead of ocamlprof.dump.

The dump file is written only if the program terminates normally (by calling exit or by falling through). It is not written if the program terminates with an uncaught exception.

If a compatible dump file already exists in the current directory, then the profiling information is accumulated in this dump file. This allows, for instance, the profiling of several executions of a program on different inputs. Note that dump files produced by byte-code executables (compiled with ocamlcp) are compatible with the dump files produced by native executables (compiled with ocamloptp).

21.3 Printing profiling information

The ocamlprof command produces a source listing of the program modules where execution counts have been inserted as comments. For instance,

        ocamlprof foo.ml

prints the source code for the foo module, with comments indicating how many times the functions in this module have been called. Naturally, this information is accurate only if the source file has not been modified after it was compiled.

The following options are recognized by ocamlprof:

-args filename
Read additional newline-terminated command line arguments from filename.
-args0 filename
Read additional null character terminated command line arguments from filename.
-f dumpfile
Specifies an alternate dump file of profiling information to be read.
-F string
Specifies an additional string to be output with profiling information. By default, ocamlprof will annotate programs with comments of the form (* n *) where n is the counter value for a profiling point. With option -F s, the annotation will be (* sn *).
-impl filename
Process the file filename as an implementation file, even if its extension is not .ml.
-intf filename
Process the file filename as an interface file, even if its extension is not .mli.
-version
Print version string and exit.
-vnum
Print short version number and exit.
-help or --help
Display a short usage summary and exit.

21.4 Time profiling

Profiling with ocamlprof only records execution counts, not the actual time spent within each function. There is currently no way to perform time profiling on bytecode programs generated by ocamlc. For time profiling of native code, users are recommended to use standard tools such as perf (on Linux), Instruments (on macOS) and DTrace. Profiling with gprof is no longer supported.

Chapter 22 Interfacing C with OCaml

This chapter describes how user-defined primitives, written in C, can be linked with OCaml code and called from OCaml functions, and how these C functions can call back to OCaml code.

22.1 Overview and compilation information

22.1.1 Declaring primitives

definition::= ...
 âˆŁâ€„external value-name : typexpr = external-declaration
 
external-declaration::= string-literal [ string-literal [ string-literal ] ]

User primitives are declared in an implementation file or struct
end module expression using the external keyword:

        external name : type = C-function-name

This defines the value name name as a function with type type that executes by calling the given C function. For instance, here is how the seek_in primitive is declared in the standard library module Stdlib:

        external seek_in : in_channel -> int -> unit = "caml_ml_seek_in"

Primitives with several arguments are always curried. The C function does not necessarily have the same name as the ML function.

External functions thus defined can be specified in interface files or sig
end signatures either as regular values

        val name : type

thus hiding their implementation as C functions, or explicitly as “manifest” external functions

        external name : type = C-function-name

The latter is slightly more efficient, as it allows clients of the module to call directly the C function instead of going through the corresponding OCaml function. On the other hand, it should not be used in library modules if they have side-effects at toplevel, as this direct call interferes with the linker’s algorithm for removing unused modules from libraries at link-time.

The arity (number of arguments) of a primitive is automatically determined from its OCaml type in the external declaration, by counting the number of function arrows in the type. For instance, seek_in above has arity 2, and the caml_ml_seek_in C function is called with two arguments. Similarly,

    external seek_in_pair: in_channel * int -> unit = "caml_ml_seek_in_pair"

has arity 1, and the caml_ml_seek_in_pair C function receives one argument (which is a pair of OCaml values).

Type abbreviations are not expanded when determining the arity of a primitive. For instance,

        type int_endo = int -> int
        external f : int_endo -> int_endo = "f"
        external g : (int -> int) -> (int -> int) = "f"

f has arity 1, but g has arity 2. This allows a primitive to return a functional value (as in the f example above): just remember to name the functional return type in a type abbreviation.

The language accepts external declarations with one or two flag strings in addition to the C function’s name. These flags are reserved for the implementation of the standard library.

22.1.2 Implementing primitives

User primitives with arity n ≀ 5 are implemented by C functions that take n arguments of type value, and return a result of type value. The type value is the type of the representations for OCaml values. It encodes objects of several base types (integers, floating-point numbers, strings, 
) as well as OCaml data structures. The type value and the associated conversion functions and macros are described in detail below. For instance, here is the declaration for the C function implementing the In_channel.input primitive, which takes 4 arguments:

CAMLprim value input(value channel, value buffer, value offset, value length)
{
  ...
}

When the primitive function is applied in an OCaml program, the C function is called with the values of the expressions to which the primitive is applied as arguments. The value returned by the function is passed back to the OCaml program as the result of the function application.

User primitives with arity greater than 5 should be implemented by two C functions. The first function, to be used in conjunction with the bytecode compiler ocamlc, receives two arguments: a pointer to an array of OCaml values (the values for the arguments), and an integer which is the number of arguments provided. The other function, to be used in conjunction with the native-code compiler ocamlopt, takes its arguments directly. For instance, here are the two C functions for the 7-argument primitive Nat.add_nat:

CAMLprim value add_nat_native(value nat1, value ofs1, value len1,
                              value nat2, value ofs2, value len2,
                              value carry_in)
{
  ...
}
CAMLprim value add_nat_bytecode(value * argv, int argn)
{
  return add_nat_native(argv[0], argv[1], argv[2], argv[3],
                        argv[4], argv[5], argv[6]);
}

The names of the two C functions must be given in the primitive declaration, as follows:

        external name : type =
                 bytecode-C-function-name native-code-C-function-name

For instance, in the case of add_nat, the declaration is:

        external add_nat: nat -> int -> int -> nat -> int -> int -> int -> int
                        = "add_nat_bytecode" "add_nat_native"

Implementing a user primitive is actually two separate tasks: on the one hand, decoding the arguments to extract C values from the given OCaml values, and encoding the return value as an OCaml value; on the other hand, actually computing the result from the arguments. Except for very simple primitives, it is often preferable to have two distinct C functions to implement these two tasks. The first function actually implements the primitive, taking native C values as arguments and returning a native C value. The second function, often called the “stub code”, is a simple wrapper around the first function that converts its arguments from OCaml values to C values, calls the first function, and converts the returned C value to an OCaml value. For instance, here is the stub code for the Int64.float_of_bits primitive:

CAMLprim value caml_int64_float_of_bits(value vi)
{
  return caml_copy_double(caml_int64_float_of_bits_unboxed(Int64_val(vi)));
}

(Here, caml_copy_double and Int64_val are conversion functions and macros for the type value, that will be described later. The CAMLprim macro expands to the required compiler directives to ensure that the function is exported and accessible from OCaml.) The hard work is performed by the function caml_int64_float_of_bits_unboxed, which is declared as:

double caml_int64_float_of_bits_unboxed(int64_t i)
{
  ...
}

To write C code that operates on OCaml values, the following include files are provided:

Include fileProvides
caml/mlvalues.hdefinition of the value type, and conversion macros
caml/alloc.hallocation functions (to create structured OCaml objects)
caml/memory.hmiscellaneous memory-related functions and macros (for GC interface, in-place modification of structures, etc).
caml/fail.hfunctions for raising exceptions (see section 22.4.5)
caml/callback.hcallback from C to OCaml (see section 22.7).
caml/custom.hoperations on custom blocks (see section 22.9).
caml/intext.hoperations for writing user-defined serialization and deserialization functions for custom blocks (see section 22.9).
caml/threads.hoperations for interfacing in the presence of multiple threads (see section 22.12).

These files reside in the caml/ subdirectory of the OCaml standard library directory, which is returned by the command ocamlc -where (usually /usr/local/lib/ocaml or /usr/lib/ocaml).

The OCaml runtime system comprises three main parts: the bytecode interpreter, the memory manager, and a set of C functions that implement the primitive operations. Some bytecode instructions are provided to call these C functions, designated by their offset in a table of functions (the table of primitives).

In the default mode, the OCaml linker produces bytecode for the standard runtime system, with a standard set of primitives. References to primitives that are not in this standard set result in the “unavailable C primitive” error. (Unless dynamic loading of C libraries is supported – see section 22.1.4 below.)

In the “custom runtime” mode, the OCaml linker scans the object files and determines the set of required primitives. Then, it builds a suitable runtime system, by calling the native code linker with:

This builds a runtime system with the required primitives. The OCaml linker generates bytecode for this custom runtime system. The bytecode is appended to the end of the custom runtime system, so that it will be automatically executed when the output file (custom runtime + bytecode) is launched.

To link in “custom runtime” mode, execute the ocamlc command with:

If you are using the native-code compiler ocamlopt, the -custom flag is not needed, as the final linking phase of ocamlopt always builds a standalone executable. To build a mixed OCaml/C executable, execute the ocamlopt command with:

Starting with Objective Caml 3.00, it is possible to record the -custom option as well as the names of C libraries in an OCaml library file .cma or .cmxa. For instance, consider an OCaml library mylib.cma, built from the OCaml object files a.cmo and b.cmo, which reference C code in libmylib.a. If the library is built as follows:

        ocamlc -a -o mylib.cma -custom a.cmo b.cmo -cclib -lmylib

users of the library can simply link with mylib.cma:

        ocamlc -o myprog mylib.cma ...

and the system will automatically add the -custom and -cclib -lmylib options, achieving the same effect as

        ocamlc -o myprog -custom a.cmo b.cmo ... -cclib -lmylib

The alternative is of course to build the library without extra options:

        ocamlc -a -o mylib.cma a.cmo b.cmo

and then ask users to provide the -custom and -cclib -lmylib options themselves at link-time:

        ocamlc -o myprog -custom mylib.cma ... -cclib -lmylib

The former alternative is more convenient for the final users of the library, however.

Starting with Objective Caml 3.03, an alternative to static linking of C code using the -custom code is provided. In this mode, the OCaml linker generates a pure bytecode executable (no embedded custom runtime system) that simply records the names of dynamically-loaded libraries containing the C code. The standard OCaml runtime system ocamlrun then loads dynamically these libraries, and resolves references to the required primitives, before executing the bytecode.

This facility is currently available on all platforms supported by OCaml except Cygwin 64 bits.

To dynamically link C code with OCaml code, the C code must first be compiled into a shared library (under Unix) or DLL (under Windows). This involves 1- compiling the C files with appropriate C compiler flags for producing position-independent code (when required by the operating system), and 2- building a shared library from the resulting object files. The resulting shared library or DLL file must be installed in a place where ocamlrun can find it later at program start-up time (see section 15.3). Finally (step 3), execute the ocamlc command with

Do not set the -custom flag, otherwise you’re back to static linking as described in section 22.1.3. The ocamlmklib tool (see section 22.14) automates steps 2 and 3.

As in the case of static linking, it is possible (and recommended) to record the names of C libraries in an OCaml .cma library archive. Consider again an OCaml library mylib.cma, built from the OCaml object files a.cmo and b.cmo, which reference C code in dllmylib.so. If the library is built as follows:

        ocamlc -a -o mylib.cma a.cmo b.cmo -dllib -lmylib

users of the library can simply link with mylib.cma:

        ocamlc -o myprog mylib.cma ...

and the system will automatically add the -dllib -lmylib option, achieving the same effect as

        ocamlc -o myprog a.cmo b.cmo ... -dllib -lmylib

Using this mechanism, users of the library mylib.cma do not need to know that it references C code, nor whether this C code must be statically linked (using -custom) or dynamically linked.

22.1.5 Choosing between static linking and dynamic linking

After having described two different ways of linking C code with OCaml code, we now review the pros and cons of each, to help developers of mixed OCaml/C libraries decide.

The main advantage of dynamic linking is that it preserves the platform-independence of bytecode executables. That is, the bytecode executable contains no machine code, and can therefore be compiled on platform A and executed on other platforms B, C, 
, as long as the required shared libraries are available on all these platforms. In contrast, executables generated by ocamlc -custom run only on the platform on which they were created, because they embark a custom-tailored runtime system specific to that platform. In addition, dynamic linking results in smaller executables.

Another advantage of dynamic linking is that the final users of the library do not need to have a C compiler, C linker, and C runtime libraries installed on their machines. This is no big deal under Unix and Cygwin, but many Windows users are reluctant to install Microsoft Visual C just to be able to do ocamlc -custom.

There are two drawbacks to dynamic linking. The first is that the resulting executable is not stand-alone: it requires the shared libraries, as well as ocamlrun, to be installed on the machine executing the code. If you wish to distribute a stand-alone executable, it is better to link it statically, using ocamlc -custom -ccopt -static or ocamlopt -ccopt -static. Dynamic linking also raises the “DLL hell” problem: some care must be taken to ensure that the right versions of the shared libraries are found at start-up time.

The second drawback of dynamic linking is that it complicates the construction of the library. The C compiler and linker flags to compile to position-independent code and build a shared library vary wildly between different Unix systems. Also, dynamic linking is not supported on all Unix systems, requiring a fall-back case to static linking in the Makefile for the library. The ocamlmklib command (see section 22.14) tries to hide some of these system dependencies.

In conclusion: dynamic linking is highly recommended under the native Windows port, because there are no portability problems and it is much more convenient for the end users. Under Unix, dynamic linking should be considered for mature, frequently used libraries because it enhances platform-independence of bytecode executables. For new or rarely-used libraries, static linking is much simpler to set up in a portable way.

22.1.6 Building standalone custom runtime systems

It is sometimes inconvenient to build a custom runtime system each time OCaml code is linked with C libraries, like ocamlc -custom does. For one thing, the building of the runtime system is slow on some systems (that have bad linkers or slow remote file systems); for another thing, the platform-independence of bytecode files is lost, forcing to perform one ocamlc -custom link per platform of interest.

An alternative to ocamlc -custom is to build separately a custom runtime system integrating the desired C libraries, then generate “pure” bytecode executables (not containing their own runtime system) that can run on this custom runtime. This is achieved by the -make-runtime and -use-runtime flags to ocamlc. For example, to build a custom runtime system integrating the C parts of the “Unix” and “Threads” libraries, do:

        ocamlc -make-runtime -o /home/me/ocamlunixrun unix.cma threads.cma

To generate a bytecode executable that runs on this runtime system, do:

        ocamlc -use-runtime /home/me/ocamlunixrun -o myprog \
                unix.cma threads.cma your .cmo and .cma files

The bytecode executable myprog can then be launched as usual: myprog args or /home/me/ocamlunixrun myprog args.

Notice that the bytecode libraries unix.cma and threads.cma must be given twice: when building the runtime system (so that ocamlc knows which C primitives are required) and also when building the bytecode executable (so that the bytecode from unix.cma and threads.cma is actually linked in).

22.2 The value type

All OCaml objects are represented by the C type value, defined in the include file caml/mlvalues.h, along with macros to manipulate values of that type. An object of type value is either:

22.2.1 Integer values

Integer values encode 63-bit signed integers (31-bit on 32-bit architectures). They are unboxed (unallocated).

22.2.2 Blocks

Blocks in the heap are garbage-collected, and therefore have strict structure constraints. Each block includes a header containing the size of the block (in words), and the tag of the block. The tag governs how the contents of the blocks are structured. A tag lower than No_scan_tag indicates a structured block, containing well-formed values, which is recursively traversed by the garbage collector. A tag greater than or equal to No_scan_tag indicates a raw block, whose contents are not scanned by the garbage collector. For the benefit of ad-hoc polymorphic primitives such as equality and structured input-output, structured and raw blocks are further classified according to their tags as follows:

TagContents of the block
0 to No_scan_tag−1A structured block (an array of OCaml objects). Each field is a value.
Closure_tagA closure representing a functional value. The first word is a pointer to a piece of code, the remaining words are value containing the environment.
String_tagA character string or a byte sequence.
Double_tagA double-precision floating-point number.
Double_array_tagAn array or record of double-precision floating-point numbers.
Abstract_tagA block representing an abstract datatype.
Custom_tagA block representing an abstract datatype with user-defined finalization, comparison, hashing, serialization and deserialization functions attached.

22.2.3 Pointers outside the heap

In earlier versions of OCaml, it was possible to use word-aligned pointers to addresses outside the heap as OCaml values, just by casting the pointer to type value. This usage is no longer supported since OCaml 5.0.

A correct way to manipulate pointers to out-of-heap blocks from OCaml is to store those pointers in OCaml blocks with tag Abstract_tag or Custom_tag, then use the blocks as the OCaml values.

Here is an example of encapsulation of out-of-heap pointers of C type ty * inside Abstract_tag blocks. Section 22.6 gives a more complete example using Custom_tag blocks.

/* Create an OCaml value encapsulating the pointer p */
static value val_of_typtr(ty * p)
{
  value v = caml_alloc(1, Abstract_tag);
  *((ty **) Data_abstract_val(v)) = p;
  return v;
}

/* Extract the pointer encapsulated in the given OCaml value */
static ty * typtr_of_val(value v)
{
  return *((ty **) Data_abstract_val(v));
}

Alternatively, out-of-heap pointers can be treated as “native” integers, that is, boxed 32-bit integers on a 32-bit platform and boxed 64-bit integers on a 64-bit platform.

/* Create an OCaml value encapsulating the pointer p */
static value val_of_typtr(ty * p)
{
  return caml_copy_nativeint((intnat) p);
}

/* Extract the pointer encapsulated in the given OCaml value */
static ty * typtr_of_val(value v)
{
  return (ty *) Nativeint_val(v);
}

For pointers that are at least 2-aligned (the low bit is guaranteed to be zero), we have yet another valid representation as an OCaml tagged integer.

/* Create an OCaml value encapsulating the pointer p */
static value val_of_typtr(ty * p)
{
  assert (((uintptr_t) p & 1) == 0);  /* check correct alignment */
  return (value) p | 1;
}

/* Extract the pointer encapsulated in the given OCaml value */
static ty * typtr_of_val(value v)
{
  return (ty *) (v & ~1);
}

22.3 Representation of OCaml data types

This section describes how OCaml data types are encoded in the value type.

22.3.1 Atomic types

OCaml typeEncoding
intUnboxed integer values.
charUnboxed integer values (ASCII code).
floatBlocks with tag Double_tag.
bytesBlocks with tag String_tag.
stringBlocks with tag String_tag.
int32Blocks with tag Custom_tag.
int64Blocks with tag Custom_tag.
nativeintBlocks with tag Custom_tag.

22.3.2 Tuples and records

Tuples are represented by pointers to blocks, with tag 0.

Records are also represented by zero-tagged blocks. The ordering of labels in the record type declaration determines the layout of the record fields: the value associated to the label declared first is stored in field 0 of the block, the value associated to the second label goes in field 1, and so on.

As an optimization, records whose fields all have static type float are represented as arrays of floating-point numbers, with tag Double_array_tag. (See the section below on arrays.)

As another optimization, unboxable record types are represented specially; unboxable record types are the immutable record types that have only one field. An unboxable type will be represented in one of two ways: boxed or unboxed. Boxed record types are represented as described above (by a block with tag 0 or Double_array_tag). An unboxed record type is represented directly by the value of its field (i.e. there is no block to represent the record itself).

The representation is chosen according to the following, in decreasing order of priority:

22.3.3 Arrays

Arrays of integers and pointers are represented like tuples and records, that is, as pointers to blocks tagged 0. They are accessed with the Field macro for reading and the caml_modify function for writing.

Values of type floatarray (as manipulated by the Float.Array module), as well as records whose declaration contains only float fields, use an efficient unboxed representation: blocks with tag Double_array_tag whose content consist of raw double values, which are not themselves valid OCaml values. They should be accessed using the Double_flat_field and Store_double_flat_field macros.

Finally, arrays of type float array may use either the boxed or the unboxed representation depending on the how the compiler is configured. They currently use the unboxed representation by default, but can be made to use the boxed representation by passing the --disable-flat-float-array flag to the ‘configure‘ script. They should be accessed using the Double_array_field and Store_double_array_field macros, which will work correctly under both modes.

22.3.4 Concrete data types

Constructed terms are represented either by unboxed integers (for constant constructors) or by blocks whose tag encode the constructor (for non-constant constructors). The constant constructors and the non-constant constructors for a given concrete type are numbered separately, starting from 0, in the order in which they appear in the concrete type declaration. A constant constructor is represented by the unboxed integer equal to its constructor number. A non-constant constructor declared with n arguments is represented by a block of size n, tagged with the constructor number; the n fields contain its arguments. Example:

Constructed termRepresentation
()Val_int(0)
falseVal_int(0)
trueVal_int(1)
[]Val_int(0)
h::tBlock with size = 2 and tag = 0; first field contains h, second field t.

As a convenience, caml/mlvalues.h defines the macros Val_unit, Val_false and Val_true to refer to (), false and true.

The following example illustrates the assignment of integers and block tags to constructors:

type t =
  | A             (* First constant constructor -> integer "Val_int(0)" *)
  | B of string   (* First non-constant constructor -> block with tag 0 *)
  | C             (* Second constant constructor -> integer "Val_int(1)" *)
  | D of bool     (* Second non-constant constructor -> block with tag 1 *)
  | E of t * t    (* Third non-constant constructor -> block with tag 2 *)

As an optimization, unboxable concrete data types are represented specially; a concrete data type is unboxable if it has exactly one constructor and this constructor has exactly one argument. Unboxable concrete data types are represented in the same ways as unboxable record types: see the description in section 22.3.2.

22.3.5 Objects

Objects are represented as blocks with tag Object_tag. The first field of the block refers to the object’s class and associated method suite, in a format that cannot easily be exploited from C. The second field contains a unique object ID, used for comparisons. The remaining fields of the object contain the values of the instance variables of the object. It is unsafe to access directly instance variables, as the type system provides no guarantee about the instance variables contained by an object.

One may extract a public method from an object using the C function caml_get_public_method (declared in <caml/mlvalues.h>.) Since public method tags are hashed in the same way as variant tags, and methods are functions taking self as first argument, if you want to do the method call foo#bar from the C side, you should call:

  callback(caml_get_public_method(foo, hash_variant("bar")), foo);

22.3.6 Polymorphic variants

Like constructed terms, polymorphic variant values are represented either as integers (for polymorphic variants without argument), or as blocks (for polymorphic variants with an argument). Unlike constructed terms, variant constructors are not numbered starting from 0, but identified by a hash value (an OCaml integer), as computed by the C function hash_variant (declared in <caml/mlvalues.h>): the hash value for a variant constructor named, say, VConstr is hash_variant("VConstr").

The variant value `VConstr is represented by hash_variant("VConstr"). The variant value `VConstr(v) is represented by a block of size 2 and tag 0, with field number 0 containing hash_variant("VConstr") and field number 1 containing v.

Unlike constructed values, polymorphic variant values taking several arguments are not flattened. That is, `VConstr(v, w) is represented by a block of size 2, whose field number 1 contains the representation of the pair (v, w), rather than a block of size 3 containing v and w in fields 1 and 2.

22.4 Operations on values

22.4.1 Kind tests

22.4.2 Operations on integers

22.4.3 Accessing blocks

The expressions Field(v, n), Byte(v, n) and Byte_u(v, n) are valid l-values. Hence, they can be assigned to, resulting in an in-place modification of value v. Assigning directly to Field(v, n) must be done with care to avoid confusing the garbage collector (see below).

22.4.4 Allocating blocks

Simple interface

Low-level interface

The following functions are slightly more efficient than caml_alloc, but also much more difficult to use.

From the standpoint of the allocation functions, blocks are divided according to their size as zero-sized blocks, small blocks (with size less than or equal to Max_young_wosize), and large blocks (with size greater than Max_young_wosize). The constant Max_young_wosize is declared in the include file mlvalues.h. It is guaranteed to be at least 64 (words), so that any block with constant size less than or equal to 64 can be assumed to be small. For blocks whose size is computed at run-time, the size must be compared against Max_young_wosize to determine the correct allocation procedure.

22.4.5 Raising exceptions

Two functions are provided to raise two standard exceptions:

Raising arbitrary exceptions from C is more delicate: the exception identifier is dynamically allocated by the OCaml program, and therefore must be communicated to the C function using the registration facility described below in section 22.7.3. Once the exception identifier is recovered in C, the following functions actually raise the exception:

22.5 Living in harmony with the garbage collector

Unused blocks in the heap are automatically reclaimed by the garbage collector. This requires some cooperation from C code that manipulates heap-allocated blocks.

22.5.1 Simple interface

All the macros described in this section are declared in the memory.h header file.

Rule 1  A function that has parameters or local variables of type value must begin with a call to one of the CAMLparam macros and return with CAMLreturn, CAMLreturn0, or CAMLreturnT.

There are six CAMLparam macros: CAMLparam0 to CAMLparam5, which take zero to five arguments respectively. If your function has no more than 5 parameters of type value, use the corresponding macros with these parameters as arguments. If your function has more than 5 parameters of type value, use CAMLparam5 with five of these parameters, and use one or more calls to the CAMLxparam macros for the remaining parameters (CAMLxparam1 to CAMLxparam5).

The macros CAMLreturn, CAMLreturn0, and CAMLreturnT are used to replace the C keyword return; any function using a CAMLparam macro on entry should use CAMLreturn on all exit points. C functions exported as OCaml externals must return a value, and they should use CAMLreturn (x) instead of return x. Some helper functions may manipulate OCaml values yet return void or another datatype. void-returning procedures should explicitly use CAMLreturn0, and not have any implicit return. Helper functions returning C data of some type t should use CAMLreturnT (t, x) instead of return x.

Note:

Some C compilers give bogus warnings about unused variables caml__dummy_xxx at each use of CAMLparam and CAMLlocal. You should ignore them.


Examples:

CAMLprim value my_external (value v1, value v2, value v3)
{
  CAMLparam3 (v1, v2, v3);
  ...
  CAMLreturn (Val_unit);
}


static void helper_procedure (value v1, value v2)
{
  CAMLparam2 (v1, v2);
  ...
  CAMLreturn0;
}

static int helper_function (value v1, value v2)
{
  CAMLparam2 (v1, v2);
  ...
  CAMLreturnT (int, 0);
}
Note:

If your function is a primitive with more than 5 arguments for use with the byte-code runtime, its arguments are not values and must not be declared (they have types value * and int).

Warning:

CAMLreturn0 should only be used for internal procedures that return void. CAMLreturn(Val_unit) should be used for functions that return an OCaml unit value. Primitives (C functions that can be called from OCaml) should never return void.

Rule 2  Local variables of type value must be declared with one of the CAMLlocal macros. Arrays of values are declared with CAMLlocalN. These macros must be used at the beginning of the function, not in a nested block.

The macros CAMLlocal1 to CAMLlocal5 declare and initialize one to five local variables of type value. The variable names are given as arguments to the macros. CAMLlocalN(x, n) declares and initializes a local variable of type value [n]. You can use several calls to these macros if you have more than 5 local variables.

Example:

CAMLprim value bar (value v1, value v2, value v3)
{
  CAMLparam3 (v1, v2, v3);
  CAMLlocal1 (result);
  result = caml_alloc (3, 0);
  ...
  CAMLreturn (result);
}
Warning:

CAMLlocal (and CAMLxparam) can only be called after CAMLparam. If a function declares local values but takes no value argument, it should start with CAMLparam0 ().

static value foo (int n)
{
  CAMLparam0 ();;
  CAMLlocal (result);
  ...
  CAMLreturn (result);
}
Rule 3  Assignments to the fields of structured blocks must be done with the Store_field macro (for normal blocks), Store_double_array_field macro (for float array values) or Store_double_flat_field (for floatarray values and records of floating-point numbers). Other assignments must not use Store_field, Store_double_array_field nor Store_double_flat_field.

Store_field (b, n, v) stores the value v in the field number n of value b, which must be a block (i.e. Is_block(b) must be true).

Example:

CAMLprim value bar (value v1, value v2, value v3)
{
  CAMLparam3 (v1, v2, v3);
  CAMLlocal1 (result);
  result = caml_alloc (3, 0);
  Store_field (result, 0, v1);
  Store_field (result, 1, v2);
  Store_field (result, 2, v3);
  CAMLreturn (result);
}
Warning:

The first argument of Store_field and Store_double_field must be a variable declared by CAMLparam* or a parameter declared by CAMLlocal* to ensure that a garbage collection triggered by the evaluation of the other arguments will not invalidate the first argument after it is computed.

Use with CAMLlocalN:

Arrays of values declared using CAMLlocalN must not be written to using Store_field. Use the normal C array syntax instead.

Rule 4  Global variables containing values must be registered with the garbage collector using the caml_register_global_root function, save that global variables and locations that will only ever contain OCaml integers (and never pointers) do not have to be registered.

The same is true for any memory location outside the OCaml heap that contains a value and is not guaranteed to be reachable—for as long as it contains such value—from either another registered global variable or location, local variable declared with CAMLlocal or function parameter declared with CAMLparam.

Registration of a global variable v is achieved by calling caml_register_global_root(&v) just before or just after a valid value is stored in v for the first time; likewise, registration of an arbitrary location p is achieved by calling caml_register_global_root(p).

You must not call any of the OCaml runtime functions or macros between registering and storing the value. Neither must you store anything in the variable v (likewise, the location p) that is not a valid value.

The registration causes the contents of the variable or memory location to be updated by the garbage collector whenever the value in such variable or location is moved within the OCaml heap. In the presence of threads care must be taken to ensure appropriate synchronisation with the OCaml runtime to avoid a race condition against the garbage collector when reading or writing the value. (See section 22.12.2.)

A registered global variable v can be un-registered by calling caml_remove_global_root(&v).

If the contents of the global variable v are seldom modified after registration, better performance can be achieved by calling caml_register_generational_global_root(&v) to register v (after its initialization with a valid value, but before any allocation or call to the GC functions), and caml_remove_generational_global_root(&v) to un-register it. In this case, you must not modify the value of v directly, but you must use caml_modify_generational_global_root(&v,x) to set it to x. The garbage collector takes advantage of the guarantee that v is not modified between calls to caml_modify_generational_global_root to scan it less often. This improves performance if the modifications of v happen less often than minor collections.

Note:

The CAML macros use identifiers (local variables, type identifiers, structure tags) that start with caml__. Do not use any identifier starting with caml__ in your programs.

22.5.2 Low-level interface

We now give the GC rules corresponding to the low-level allocation functions caml_alloc_small and caml_alloc_shr. You can ignore those rules if you stick to the simplified allocation function caml_alloc.

Rule 5  After a structured block (a block with tag less than No_scan_tag) is allocated with the low-level functions, all fields of this block must be filled with well-formed values before the next allocation operation. If the block has been allocated with caml_alloc_small, filling is performed by direct assignment to the fields of the block:

        Field(v, n) = vn;
If the block has been allocated with caml_alloc_shr, filling is performed through the caml_initialize function:

        caml_initialize(&Field(v, n), vn);

The next allocation can trigger a garbage collection. The garbage collector assumes that all structured blocks contain well-formed values. Newly created blocks contain random data, which generally do not represent well-formed values.

If you really need to allocate before the fields can receive their final value, first initialize with a constant value (e.g. Val_unit), then allocate, then modify the fields with the correct value (see rule 6).

Rule 6  Direct assignment to a field of a block, as in

        Field(v, n) = w;
is safe only if v is a block newly allocated by caml_alloc_small; that is, if no allocation took place between the allocation of v and the assignment to the field. In all other cases, never assign directly. If the block has just been allocated by caml_alloc_shr, use caml_initialize to assign a value to a field for the first time:

        caml_initialize(&Field(v, n), w);
Otherwise, you are updating a field that previously contained a well-formed value; then, call the caml_modify function:

        caml_modify(&Field(v, n), w);

To illustrate the rules above, here is a C function that builds and returns a list containing the two integers given as parameters. First, we write it using the simplified allocation functions:

value alloc_list_int(int i1, int i2)
{
  CAMLparam0 ();
  CAMLlocal2 (result, r);

  r = caml_alloc(2, 0);                   /* Allocate a cons cell */
  Store_field(r, 0, Val_int(i2));         /* car = the integer i2 */
  Store_field(r, 1, Val_int(0));          /* cdr = the empty list [] */
  result = caml_alloc(2, 0);              /* Allocate the other cons cell */
  Store_field(result, 0, Val_int(i1));    /* car = the integer i1 */
  Store_field(result, 1, r);              /* cdr = the first cons cell */
  CAMLreturn (result);
}

Here, the registering of result is not strictly needed, because no allocation takes place after it gets its value, but it’s easier and safer to simply register all the local variables that have type value.

Here is the same function written using the low-level allocation functions. We notice that the cons cells are small blocks and can be allocated with caml_alloc_small, and filled by direct assignments on their fields.

value alloc_list_int(int i1, int i2)
{
  CAMLparam0 ();
  CAMLlocal2 (result, r);

  r = caml_alloc_small(2, 0);             /* Allocate a cons cell */
  Field(r, 0) = Val_int(i2);              /* car = the integer i2 */
  Field(r, 1) = Val_int(0);               /* cdr = the empty list [] */
  result = caml_alloc_small(2, 0);        /* Allocate the other cons cell */
  Field(result, 0) = Val_int(i1);         /* car = the integer i1 */
  Field(result, 1) = r;                   /* cdr = the first cons cell */
  CAMLreturn (result);
}

In the two examples above, the list is built bottom-up. Here is an alternate way, that proceeds top-down. It is less efficient, but illustrates the use of caml_modify.

value alloc_list_int(int i1, int i2)
{
  CAMLparam0 ();
  CAMLlocal2 (tail, r);

  r = caml_alloc_small(2, 0);             /* Allocate a cons cell */
  Field(r, 0) = Val_int(i1);              /* car = the integer i1 */
  Field(r, 1) = Val_int(0);               /* A dummy value
  tail = caml_alloc_small(2, 0);          /* Allocate the other cons cell */
  Field(tail, 0) = Val_int(i2);           /* car = the integer i2 */
  Field(tail, 1) = Val_int(0);            /* cdr = the empty list [] */
  caml_modify(&Field(r, 1), tail);        /* cdr of the result = tail */
  CAMLreturn (r);
}

It would be incorrect to perform Field(r, 1) = tail directly, because the allocation of tail has taken place since r was allocated.

22.5.3 Pending actions and asynchronous exceptions

Since 4.10, allocation functions are guaranteed not to call any OCaml callbacks from C, including finalisers and signal handlers, and delay their execution instead.

The function caml_process_pending_actions from <caml/signals.h> executes any pending signal handlers and finalisers, Memprof callbacks, and requested minor and major garbage collections. In particular, it can raise asynchronous exceptions. It is recommended to call it regularly at safe points inside long-running non-blocking C code.

The variant caml_process_pending_actions_exn is provided, that returns the exception instead of raising it directly into OCaml code. Its result must be tested using Is_exception_result, and followed by Extract_exception if appropriate. It is typically used for clean up before re-raising:

    CAMLlocal1(exn);
    ...
    exn = caml_process_pending_actions_exn();
    if(Is_exception_result(exn)) {
      exn = Extract_exception(exn);
      ...cleanup...
      caml_raise(exn);
    }

Correct use of exceptional return, in particular in the presence of garbage collection, is further detailed in Section 22.7.1.

22.6 A complete example

This section outlines how the functions from the Unix curses library can be made available to OCaml programs. First of all, here is the interface curses.ml that declares the curses primitives and data types:

(* File curses.ml -- declaration of primitives and data types *)
type window                   (* The type "window" remains abstract *)
external initscr: unit -> window = "caml_curses_initscr"
external endwin: unit -> unit = "caml_curses_endwin"
external refresh: unit -> unit = "caml_curses_refresh"
external wrefresh : window -> unit = "caml_curses_wrefresh"
external newwin: int -> int -> int -> int -> window = "caml_curses_newwin"
external addch: char -> unit = "caml_curses_addch"
external mvwaddch: window -> int -> int -> char -> unit = "caml_curses_mvwaddch"
external addstr: string -> unit = "caml_curses_addstr"
external mvwaddstr: window -> int -> int -> string -> unit
         = "caml_curses_mvwaddstr"
(* lots more omitted *)

To compile this interface:

        ocamlc -c curses.ml

To implement these functions, we just have to provide the stub code; the core functions are already implemented in the curses library. The stub code file, curses_stubs.c, looks like this:

/* File curses_stubs.c -- stub code for curses */
#include <curses.h>
#include <caml/mlvalues.h>
#include <caml/memory.h>
#include <caml/alloc.h>
#include <caml/custom.h>

/* Encapsulation of opaque window handles (of type WINDOW *)
   as OCaml custom blocks. */

static struct custom_operations curses_window_ops = {
  "fr.inria.caml.curses_windows",
  custom_finalize_default,
  custom_compare_default,
  custom_hash_default,
  custom_serialize_default,
  custom_deserialize_default,
  custom_compare_ext_default,
  custom_fixed_length_default
};

/* Accessing the WINDOW * part of an OCaml custom block */
#define Window_val(v) (*((WINDOW **) Data_custom_val(v)))

/* Allocating an OCaml custom block to hold the given WINDOW * */
static value alloc_window(WINDOW * w)
{
  value v = caml_alloc_custom(&curses_window_ops, sizeof(WINDOW *), 0, 1);
  Window_val(v) = w;
  return v;
}

CAMLprim value caml_curses_initscr(value unit)
{
  CAMLparam1 (unit);
  CAMLreturn (alloc_window(initscr()));
}

CAMLprim value caml_curses_endwin(value unit)
{
  CAMLparam1 (unit);
  endwin();
  CAMLreturn (Val_unit);
}

CAMLprim value caml_curses_refresh(value unit)
{
  CAMLparam1 (unit);
  refresh();
  CAMLreturn (Val_unit);
}

CAMLprim value caml_curses_wrefresh(value win)
{
  CAMLparam1 (win);
  wrefresh(Window_val(win));
  CAMLreturn (Val_unit);
}

CAMLprim value caml_curses_newwin(value nlines, value ncols, value x0, value y0)
{
  CAMLparam4 (nlines, ncols, x0, y0);
  CAMLreturn (alloc_window(newwin(Int_val(nlines), Int_val(ncols),
                                  Int_val(x0), Int_val(y0))));
}

CAMLprim value caml_curses_addch(value c)
{
  CAMLparam1 (c);
  addch(Int_val(c));            /* Characters are encoded like integers */
  CAMLreturn (Val_unit);
}

CAMLprim value caml_curses_mvwaddch(value win, value x, value y, value c)
{
  CAMLparam4 (win, x, y, c);
  mvwaddch(Window_val(win), Int_val(x), Int_val(y), Int_val(c));
  CAMLreturn (Val_unit);
}

CAMLprim value caml_curses_addstr(value s)
{
  CAMLparam1 (s);
  addstr(String_val(s));
  CAMLreturn (Val_unit);
}

CAMLprim value caml_curses_mvwaddstr(value win, value x, value y, value s)
{
  CAMLparam4 (win, x, y, s);
  mvwaddstr(Window_val(win), Int_val(x), Int_val(y), String_val(s));
  CAMLreturn (Val_unit);
}

/* This goes on for pages. */

The file curses_stubs.c can be compiled with:

        cc -c -I`ocamlc -where` curses_stubs.c

or, even simpler,

        ocamlc -c curses_stubs.c

(When passed a .c file, the ocamlc command simply calls the C compiler on that file, with the right -I option.)

Now, here is a sample OCaml program prog.ml that uses the curses module:

(* File prog.ml -- main program using curses *)
open Curses;;
let main_window = initscr () in
let small_window = newwin 10 5 20 10 in
  mvwaddstr main_window 10 2 "Hello";
  mvwaddstr small_window 4 3 "world";
  refresh();
  Unix.sleep 5;
  endwin()

To compile and link this program, run:

       ocamlc -custom -o prog unix.cma curses.cmo prog.ml curses_stubs.o -cclib -lcurses

(On some machines, you may need to put -cclib -lcurses -cclib -ltermcap or -cclib -ltermcap instead of -cclib -lcurses.)

22.7 Advanced topic: callbacks from C to OCaml

So far, we have described how to call C functions from OCaml. In this section, we show how C functions can call OCaml functions, either as callbacks (OCaml calls C which calls OCaml), or with the main program written in C.

22.7.1 Applying OCaml closures from C

C functions can apply OCaml function values (closures) to OCaml values. The following functions are provided to perform the applications:

If the function f does not return, but raises an exception that escapes the scope of the application, then this exception is propagated to the next enclosing OCaml code, skipping over the C code. That is, if an OCaml function f calls a C function g that calls back an OCaml function h that raises a stray exception, then the execution of g is interrupted and the exception is propagated back into f.

If the C code wishes to catch exceptions escaping the OCaml function, it can use the functions caml_callback_exn, caml_callback2_exn, caml_callback3_exn, caml_callbackN_exn. These functions take the same arguments as their non-_exn counterparts, but catch escaping exceptions and return them to the C code. The return value v of the caml_callback*_exn functions must be tested with the macro Is_exception_result(v). If the macro returns “false”, no exception occurred, and v is the value returned by the OCaml function. If Is_exception_result(v) returns “true”, an exception escaped, and its value (the exception descriptor) can be recovered using Extract_exception(v).

Warning:

If the OCaml function returned with an exception, Extract_exception should be applied to the exception result prior to calling a function that may trigger garbage collection. Otherwise, if v is reachable during garbage collection, the runtime can crash since v does not contain a valid value.

Example:

    CAMLprim value call_caml_f_ex(value closure, value arg)
    {
      CAMLparam2(closure, arg);
      CAMLlocal2(res, tmp);
      res = caml_callback_exn(closure, arg);
      if(Is_exception_result(res)) {
        res = Extract_exception(res);
        tmp = caml_alloc(3, 0); /* Safe to allocate: res contains valid value. */
        ...
      }
      CAMLreturn (res);
    }

22.7.2 Obtaining or registering OCaml closures for use in C functions

There are two ways to obtain OCaml function values (closures) to be passed to the callback functions described above. One way is to pass the OCaml function as an argument to a primitive function. For example, if the OCaml code contains the declaration

    external apply : ('a -> 'b) -> 'a -> 'b = "caml_apply"

the corresponding C stub can be written as follows:

    CAMLprim value caml_apply(value vf, value vx)
    {
      CAMLparam2(vf, vx);
      CAMLlocal1(vy);
      vy = caml_callback(vf, vx);
      CAMLreturn(vy);
    }

Another possibility is to use the registration mechanism provided by OCaml. This registration mechanism enables OCaml code to register OCaml functions under some global name, and C code to retrieve the corresponding closure by this global name.

On the OCaml side, registration is performed by evaluating Callback.register n v. Here, n is the global name (an arbitrary string) and v the OCaml value. For instance:

    let f x = print_string "f is applied to "; print_int x; print_newline()
    let _ = Callback.register "test function" f

On the C side, a pointer to the value registered under name n is obtained by calling caml_named_value(n). The returned pointer must then be dereferenced to recover the actual OCaml value. If no value is registered under the name n, the null pointer is returned. For example, here is a C wrapper that calls the OCaml function f above:

    void call_caml_f(int arg)
    {
        caml_callback(*caml_named_value("test function"), Val_int(arg));
    }

The pointer returned by caml_named_value is constant and can safely be cached in a C variable to avoid repeated name lookups. The value pointed to cannot be changed from C. However, it might change during garbage collection, so must always be recomputed at the point of use. Here is a more efficient variant of call_caml_f above that calls caml_named_value only once:

    void call_caml_f(int arg)
    {
        static const value * closure_f = NULL;
        if (closure_f == NULL) {
            /* First time around, look up by name */
            closure_f = caml_named_value("test function");
        }
        caml_callback(*closure_f, Val_int(arg));
    }

22.7.3 Registering OCaml exceptions for use in C functions

The registration mechanism described above can also be used to communicate exception identifiers from OCaml to C. The OCaml code registers the exception by evaluating Callback.register_exception n exn, where n is an arbitrary name and exn is an exception value of the exception to register. For example:

    exception Error of string
    let _ = Callback.register_exception "test exception" (Error "any string")

The C code can then recover the exception identifier using caml_named_value and pass it as first argument to the functions raise_constant, raise_with_arg, and raise_with_string (described in section 22.4.5) to actually raise the exception. For example, here is a C function that raises the Error exception with the given argument:

    void raise_error(char * msg)
    {
        caml_raise_with_string(*caml_named_value("test exception"), msg);
    }

22.7.4 Main program in C

In normal operation, a mixed OCaml/C program starts by executing the OCaml initialization code, which then may proceed to call C functions. We say that the main program is the OCaml code. In some applications, it is desirable that the C code plays the role of the main program, calling OCaml functions when needed. This can be achieved as follows:

22.7.5 Embedding the OCaml code in the C code

The bytecode compiler in custom runtime mode (ocamlc -custom) normally appends the bytecode to the executable file containing the custom runtime. This has two consequences. First, the final linking step must be performed by ocamlc. Second, the OCaml runtime library must be able to find the name of the executable file from the command-line arguments. When using caml_main(argv) as in section 22.7.4, this means that argv[0] or argv[1] must contain the executable file name.

An alternative is to embed the bytecode in the C code. The -output-obj and -output-complete-obj options to ocamlc are provided for this purpose. They cause the ocamlc compiler to output a C object file (.o file, .obj under Windows) containing the bytecode for the OCaml part of the program, as well as a caml_startup function. The C object file produced by ocamlc -output-complete-obj also contains the runtime and autolink libraries. The C object file produced by ocamlc -output-obj or ocamlc -output-complete-obj can then be linked with C code using the standard C compiler, or stored in a C library.

The caml_startup function must be called from the main C program in order to initialize the OCaml runtime and execute the OCaml initialization code. Just like caml_main, it takes one argv parameter containing the command-line parameters. Unlike caml_main, this argv parameter is used only to initialize Sys.argv, but not for finding the name of the executable file.

The caml_startup function calls the uncaught exception handler (or enters the debugger, if running under ocamldebug) if an exception escapes from a top-level module initialiser. Such exceptions may be caught in the C code by instead using the caml_startup_exn function and testing the result using Is_exception_result (followed by Extract_exception if appropriate).

The -output-obj and -output-complete-obj options can also be used to obtain the C source file. More interestingly, these options can also produce directly a shared library (.so file, .dll under Windows) that contains the OCaml code, the OCaml runtime system and any other static C code given to ocamlc (.o, .a, respectively, .obj, .lib). This use of -output-obj and -output-complete-obj is very similar to a normal linking step, but instead of producing a main program that automatically runs the OCaml code, it produces a shared library that can run the OCaml code on demand. The three possible behaviors of -output-obj and -output-complete-obj (to produce a C source code .c, a C object file .o, a shared library .so), are selected according to the extension of the resulting file (given with -o).

The native-code compiler ocamlopt also supports the -output-obj and -output-complete-obj options, causing it to output a C object file or a shared library containing the native code for all OCaml modules on the command-line, as well as the OCaml startup code. Initialization is performed by calling caml_startup (or caml_startup_exn) as in the case of the bytecode compiler. The file produced by ocamlopt -output-complete-obj also contains the runtime and autolink libraries.

For the final linking phase, in addition to the object file produced by -output-obj, you will have to provide the OCaml runtime library (libcamlrun.a for bytecode, libasmrun.a for native-code), as well as all C libraries that are required by the OCaml libraries used. For instance, assume the OCaml part of your program uses the Unix library. With ocamlc, you should do:

        ocamlc -output-obj -o camlcode.o unix.cma other .cmo and .cma files
        cc -o myprog C objects and libraries \
           camlcode.o -L‘ocamlc -where‘ -lunix -lcamlrun

With ocamlopt, you should do:

        ocamlopt -output-obj -o camlcode.o unix.cmxa other .cmx and .cmxa files
        cc -o myprog C objects and libraries \
           camlcode.o -L‘ocamlc -where‘ -lunix -lasmrun

For the final linking phase, in addition to the object file produced by -output-complete-obj, you will have only to provide the C libraries required by the OCaml runtime.

For instance, assume the OCaml part of your program uses the Unix library. With ocamlc, you should do:

        ocamlc -output-complete-obj -o camlcode.o unix.cma other .cmo and .cma files
        cc -o myprog C objects and libraries \
           camlcode.o C libraries required by the runtime, eg -lm  -ldl -lcurses -lpthread

With ocamlopt, you should do:

        ocamlopt -output-complete-obj -o camlcode.o unix.cmxa other .cmx and .cmxa files
        cc -o myprog C objects and libraries \
           camlcode.o C libraries required by the runtime, eg -lm -ldl
Warning:

On some ports, special options are required on the final linking phase that links together the object file produced by the -output-obj and -output-complete-obj options and the remainder of the program. Those options are shown in the configuration file Makefile.config generated during compilation of OCaml, as the variable OC_LDFLAGS.

Stack backtraces.

When OCaml bytecode produced by ocamlc -g is embedded in a C program, no debugging information is included, and therefore it is impossible to print stack backtraces on uncaught exceptions. This is not the case when native code produced by ocamlopt -g is embedded in a C program: stack backtrace information is available, but the backtrace mechanism needs to be turned on programmatically. This can be achieved from the OCaml side by calling Printexc.record_backtrace true in the initialization of one of the OCaml modules. This can also be achieved from the C side by calling caml_record_backtraces(1); in the OCaml-C glue code. (caml_record_backtraces is declared in backtrace.h)

Unloading the runtime.

In case the shared library produced with -output-obj is to be loaded and unloaded repeatedly by a single process, care must be taken to unload the OCaml runtime explicitly, in order to avoid various system resource leaks.

Since 4.05, caml_shutdown function can be used to shut the runtime down gracefully, which equals the following:

As a shared library may have several clients simultaneously, it is made for convenience that caml_startup (and caml_startup_pooled) may be called multiple times, given that each such call is paired with a corresponding call to caml_shutdown (in a nested fashion). The runtime will be unloaded once there are no outstanding calls to caml_startup.

Once a runtime is unloaded, it cannot be started up again without reloading the shared library and reinitializing its static data. Therefore, at the moment, the facility is only useful for building reloadable shared libraries.

Unix signal handling.

Depending on the target platform and operating system, the native-code runtime system may install signal handlers for one or several of the SIGSEGV, SIGTRAP and SIGFPE signals when caml_startup is called, and reset these signals to their default behaviors when caml_shutdown is called. The main program written in C should not try to handle these signals itself.

22.8 Advanced example with callbacks

This section illustrates the callback facilities described in section 22.7. We are going to package some OCaml functions in such a way that they can be linked with C code and called from C just like any C functions. The OCaml functions are defined in the following mod.ml OCaml source:

(* File mod.ml -- some "useful" OCaml functions *)

let rec fib n = if n < 2 then 1 else fib(n-1) + fib(n-2)

let format_result n = Printf.sprintf "Result is: %d\n" n

(* Export those two functions to C *)

let _ = Callback.register "fib" fib
let _ = Callback.register "format_result" format_result

Here is the C stub code for calling these functions from C:

/* File modwrap.c -- wrappers around the OCaml functions */

#include <stdio.h>
#include <string.h>
#include <caml/mlvalues.h>
#include <caml/callback.h>

int fib(int n)
{
  static const value * fib_closure = NULL;
  if (fib_closure == NULL) fib_closure = caml_named_value("fib");
  return Int_val(caml_callback(*fib_closure, Val_int(n)));
}

char * format_result(int n)
{
  static const value * format_result_closure = NULL;
  if (format_result_closure == NULL)
    format_result_closure = caml_named_value("format_result");
  return strdup(String_val(caml_callback(*format_result_closure, Val_int(n))));
  /* We copy the C string returned by String_val to the C heap
     so that it remains valid after garbage collection. */
}

We now compile the OCaml code to a C object file and put it in a C library along with the stub code in modwrap.c and the OCaml runtime system:

        ocamlc -custom -output-obj -o modcaml.o mod.ml
        ocamlc -c modwrap.c
        cp `ocamlc -where`/libcamlrun.a mod.a && chmod +w mod.a
        ar r mod.a modcaml.o modwrap.o

(One can also use ocamlopt -output-obj instead of ocamlc -custom -output-obj. In this case, replace libcamlrun.a (the bytecode runtime library) by libasmrun.a (the native-code runtime library).)

Now, we can use the two functions fib and format_result in any C program, just like regular C functions. Just remember to call caml_startup (or caml_startup_exn) once before.

/* File main.c -- a sample client for the OCaml functions */

#include <stdio.h>
#include <caml/callback.h>

extern int fib(int n);
extern char * format_result(int n);

int main(int argc, char ** argv)
{
  int result;

  /* Initialize OCaml code */
  caml_startup(argv);
  /* Do some computation */
  result = fib(10);
  printf("fib(10) = %s\n", format_result(result));
  return 0;
}

To build the whole program, just invoke the C compiler as follows:

        cc -o prog -I `ocamlc -where` main.c mod.a -lcurses

(On some machines, you may need to put -ltermcap or -lcurses -ltermcap instead of -lcurses.)

22.9 Advanced topic: custom blocks

Blocks with tag Custom_tag contain both arbitrary user data and a pointer to a C struct, with type struct custom_operations, that associates user-provided finalization, comparison, hashing, serialization and deserialization functions to this block.

22.9.1 The struct custom_operations

The struct custom_operations is defined in <caml/custom.h> and contains the following fields:

Note: the finalize, compare, hash, serialize and deserialize functions attached to custom block descriptors must never access the OCaml runtime. Within these functions, do not call any of the OCaml allocation functions, and do not perform a callback into OCaml code. Do not use CAMLparam to register the parameters to these functions, and do not use CAMLreturn to return the result. Do not raise exceptions, do not remove global roots, etc.

22.9.2 Allocating custom blocks

Custom blocks must be allocated via caml_alloc_custom or caml_alloc_custom_mem:

caml_alloc_custom(ops, size, used, max)

returns a fresh custom block, with room for size bytes of user data, and whose associated operations are given by ops (a pointer to a struct custom_operations, usually statically allocated as a C global variable).

The two parameters used and max are used to control the speed of garbage collection when the finalized object contains pointers to out-of-heap resources. Generally speaking, the OCaml incremental major collector adjusts its speed relative to the allocation rate of the program. The faster the program allocates, the harder the GC works in order to reclaim quickly unreachable blocks and avoid having large amount of “floating garbage” (unreferenced objects that the GC has not yet collected).

Normally, the allocation rate is measured by counting the in-heap size of allocated blocks. However, it often happens that finalized objects contain pointers to out-of-heap memory blocks and other resources (such as file descriptors, X Windows bitmaps, etc.). For those blocks, the in-heap size of blocks is not a good measure of the quantity of resources allocated by the program.

The two arguments used and max give the GC an idea of how much out-of-heap resources are consumed by the finalized block being allocated: you give the amount of resources allocated to this object as parameter used, and the maximum amount that you want to see in floating garbage as parameter max. The units are arbitrary: the GC cares only about the ratio used / max.

For instance, if you are allocating a finalized block holding an X Windows bitmap of w by h pixels, and you’d rather not have more than 1 mega-pixels of unreclaimed bitmaps, specify used = w * h and max = 1000000.

Another way to describe the effect of the used and max parameters is in terms of full GC cycles. If you allocate many custom blocks with used / max = 1 / N, the GC will then do one full cycle (examining every object in the heap and calling finalization functions on those that are unreachable) every N allocations. For instance, if used = 1 and max = 1000, the GC will do one full cycle at least every 1000 allocations of custom blocks.

If your finalized blocks contain no pointers to out-of-heap resources, or if the previous discussion made little sense to you, just take used = 0 and max = 1. But if you later find that the finalization functions are not called “often enough”, consider increasing the used / max ratio.

caml_alloc_custom_mem(ops, size, used)

Use this function when your custom block holds only out-of-heap memory (memory allocated with malloc or caml_stat_alloc) and no other resources. used should be the number of bytes of out-of-heap memory that are held by your custom block. This function works like caml_alloc_custom except that the max parameter is under the control of the user (via the custom_major_ratio, custom_minor_ratio, and custom_minor_max_size parameters) and proportional to the heap sizes. It has been available since OCaml 4.08.0.

22.9.3 Accessing custom blocks

The data part of a custom block v can be accessed via the pointer Data_custom_val(v). This pointer has type void * and should be cast to the actual type of the data stored in the custom block.

The contents of custom blocks are not scanned by the garbage collector, and must therefore not contain any pointer inside the OCaml heap. In other terms, never store an OCaml value in a custom block, and do not use Field, Store_field nor caml_modify to access the data part of a custom block. Conversely, any C data structure (not containing heap pointers) can be stored in a custom block.

22.9.4 Writing custom serialization and deserialization functions

The following functions, defined in <caml/intext.h>, are provided to write and read back the contents of custom blocks in a portable way. Those functions handle endianness conversions when e.g. data is written on a little-endian machine and read back on a big-endian machine.

FunctionAction
caml_serialize_int_1Write a 1-byte integer
caml_serialize_int_2Write a 2-byte integer
caml_serialize_int_4Write a 4-byte integer
caml_serialize_int_8Write a 8-byte integer
caml_serialize_float_4Write a 4-byte float
caml_serialize_float_8Write a 8-byte float
caml_serialize_block_1Write an array of 1-byte quantities
caml_serialize_block_2Write an array of 2-byte quantities
caml_serialize_block_4Write an array of 4-byte quantities
caml_serialize_block_8Write an array of 8-byte quantities
caml_deserialize_uint_1Read an unsigned 1-byte integer
caml_deserialize_sint_1Read a signed 1-byte integer
caml_deserialize_uint_2Read an unsigned 2-byte integer
caml_deserialize_sint_2Read a signed 2-byte integer
caml_deserialize_uint_4Read an unsigned 4-byte integer
caml_deserialize_sint_4Read a signed 4-byte integer
caml_deserialize_uint_8Read an unsigned 8-byte integer
caml_deserialize_sint_8Read a signed 8-byte integer
caml_deserialize_float_4Read a 4-byte float
caml_deserialize_float_8Read an 8-byte float
caml_deserialize_block_1Read an array of 1-byte quantities
caml_deserialize_block_2Read an array of 2-byte quantities
caml_deserialize_block_4Read an array of 4-byte quantities
caml_deserialize_block_8Read an array of 8-byte quantities
caml_deserialize_errorSignal an error during deserialization; input_value or Marshal.from_... raise a Failure exception after cleaning up their internal data structures

Serialization functions are attached to the custom blocks to which they apply. Obviously, deserialization functions cannot be attached this way, since the custom block does not exist yet when deserialization begins! Thus, the struct custom_operations that contain deserialization functions must be registered with the deserializer in advance, using the register_custom_operations function declared in <caml/custom.h>. Deserialization proceeds by reading the identifier off the input stream, allocating a custom block of the size specified in the input stream, searching the registered struct custom_operation blocks for one with the same identifier, and calling its deserialize function to fill the data part of the custom block.

22.9.5 Choosing identifiers

Identifiers in struct custom_operations must be chosen carefully, since they must identify uniquely the data structure for serialization and deserialization operations. In particular, consider including a version number in the identifier; this way, the format of the data can be changed later, yet backward-compatible deserialisation functions can be provided.

Identifiers starting with _ (an underscore character) are reserved for the OCaml runtime system; do not use them for your custom data. We recommend to use a URL (http://mymachine.mydomain.com/mylibrary/version-number) or a Java-style package name (com.mydomain.mymachine.mylibrary.version-number) as identifiers, to minimize the risk of identifier collision.

22.9.6 Finalized blocks

Custom blocks generalize the finalized blocks that were present in OCaml prior to version 3.00. For backwards compatibility, the format of custom blocks is compatible with that of finalized blocks, and the caml_alloc_final function is still available to allocate a custom block with a given finalization function, but default comparison, hashing and serialization functions. (In particular, the finalization function must not access the OCaml runtime.)

caml_alloc_final(n, f, used, max) returns a fresh custom block of size n+1 words, with finalization function f. The first word is reserved for storing the custom operations; the other n words are available for your data. The two parameters used and max are used to control the speed of garbage collection, as described for caml_alloc_custom.

22.10 Advanced topic: Bigarrays and the OCaml-C interface

This section explains how C stub code that interfaces C or Fortran code with OCaml code can use Bigarrays.

22.10.1 Include file

The include file <caml/bigarray.h> must be included in the C stub file. It declares the functions, constants and macros discussed below.

22.10.2 Accessing an OCaml bigarray from C or Fortran

If v is a OCaml value representing a Bigarray, the expression Caml_ba_data_val(v) returns a pointer to the data part of the array. This pointer is of type void * and can be cast to the appropriate C type for the array (e.g. double [], char [][10], etc).

Various characteristics of the OCaml Bigarray can be consulted from C as follows:

C expressionReturns
Caml_ba_array_val(v)->num_dimsnumber of dimensions
Caml_ba_array_val(v)->dim[i]i-th dimension
Caml_ba_array_val(v)->flags & BIGARRAY_KIND_MASKkind of array elements

The kind of array elements is one of the following constants:

ConstantElement kind
CAML_BA_FLOAT3232-bit single-precision floats
CAML_BA_FLOAT6464-bit double-precision floats
CAML_BA_SINT88-bit signed integers
CAML_BA_UINT88-bit unsigned integers
CAML_BA_SINT1616-bit signed integers
CAML_BA_UINT1616-bit unsigned integers
CAML_BA_INT3232-bit signed integers
CAML_BA_INT6464-bit signed integers
CAML_BA_CAML_INT31- or 63-bit signed integers
CAML_BA_NATIVE_INT32- or 64-bit (platform-native) integers
CAML_BA_COMPLEX3232-bit single-precision complex numbers
CAML_BA_COMPLEX6464-bit double-precision complex numbers
CAML_BA_CHAR8-bit characters
Warning:

Caml_ba_array_val(v) must always be dereferenced immediately and not stored anywhere, including local variables. It resolves to a derived pointer: it is not a valid OCaml value but points to a memory region managed by the GC. For this reason this value must not be stored in any memory location that could be live cross a GC.

The following example shows the passing of a two-dimensional Bigarray to a C function and a Fortran function.

    extern void my_c_function(double * data, int dimx, int dimy);
    extern void my_fortran_function_(double * data, int * dimx, int * dimy);

    CAMLprim value caml_stub(value bigarray)
    {
      int dimx = Caml_ba_array_val(bigarray)->dim[0];
      int dimy = Caml_ba_array_val(bigarray)->dim[1];
      /* C passes scalar parameters by value */
      my_c_function(Caml_ba_data_val(bigarray), dimx, dimy);
      /* Fortran passes all parameters by reference */
      my_fortran_function_(Caml_ba_data_val(bigarray), &dimx, &dimy);
      return Val_unit;
    }

22.10.3 Wrapping a C or Fortran array as an OCaml Bigarray

A pointer p to an already-allocated C or Fortran array can be wrapped and returned to OCaml as a Bigarray using the caml_ba_alloc or caml_ba_alloc_dims functions.

The following example illustrates how statically-allocated C and Fortran arrays can be made available to OCaml.

    extern long my_c_array[100][200];
    extern float my_fortran_array_[300][400];

    CAMLprim value caml_get_c_array(value unit)
    {
      long dims[2];
      dims[0] = 100; dims[1] = 200;
      return caml_ba_alloc(CAML_BA_NATIVE_INT | CAML_BA_C_LAYOUT,
                           2, my_c_array, dims);
    }

    CAMLprim value caml_get_fortran_array(value unit)
    {
      return caml_ba_alloc_dims(CAML_BA_FLOAT32 | CAML_BA_FORTRAN_LAYOUT,
                                2, my_fortran_array_, 300L, 400L);
    }

22.11 Advanced topic: cheaper C call

This section describe how to make calling C functions cheaper.

Note: This only applies to the native compiler. So whenever you use any of these methods, you have to provide an alternative byte-code stub that ignores all the special annotations.

22.11.1 Passing unboxed values

We said earlier that all OCaml objects are represented by the C type value, and one has to use macros such as Int_val to decode data from the value type. It is however possible to tell the OCaml native-code compiler to do this for us and pass arguments unboxed to the C function. Similarly it is possible to tell OCaml to expect the result unboxed and box it for us.

The motivation is that, by letting ‘ocamlopt‘ deal with boxing, it can often decide to suppress it entirely.

For instance let’s consider this example:

external foo : float -> float -> float = "foo"

let f a b =
  let len = Array.length a in
  assert (Array.length b = len);
  let res = Array.make len 0. in
  for i = 0 to len - 1 do
    res.(i) <- foo a.(i) b.(i)
  done

Float arrays are unboxed in OCaml, however the C function foo expect its arguments as boxed floats and returns a boxed float. Hence the OCaml compiler has no choice but to box a.(i) and b.(i) and unbox the result of foo. This results in the allocation of 3 * len temporary float values.

Now if we annotate the arguments and result with [@unboxed], the native-code compiler will be able to avoid all these allocations:

external foo
  :  (float [@unboxed])
  -> (float [@unboxed])
  -> (float [@unboxed])
  = "foo_byte" "foo"

In this case the C functions must look like:

CAMLprim double foo(double a, double b)
{
  ...
}

CAMLprim value foo_byte(value a, value b)
{
  return caml_copy_double(foo(Double_val(a), Double_val(b)))
}

For convenience, when all arguments and the result are annotated with [@unboxed], it is possible to put the attribute only once on the declaration itself. So we can also write instead:

external foo : float -> float -> float = "foo_byte" "foo" [@@unboxed]

The following table summarize what OCaml types can be unboxed, and what C types should be used in correspondence:

OCaml typeC type
floatdouble
int32int32_t
int64int64_t
nativeintintnat

Similarly, it is possible to pass untagged OCaml integers between OCaml and C. This is done by annotating the arguments and/or result with [@untagged]:

external f : string -> (int [@untagged]) = "f_byte" "f"

The corresponding C type must be intnat.

Note: Do not use the C int type in correspondence with (int [@untagged]). This is because they often differ in size.

22.11.2 Direct C call

In order to be able to run the garbage collector in the middle of a C function, the OCaml native-code compiler generates some bookkeeping code around C calls. Technically it wraps every C call with the C function caml_c_call which is part of the OCaml runtime.

For small functions that are called repeatedly, this indirection can have a big impact on performances. However this is not needed if we know that the C function doesn’t allocate, doesn’t raise exceptions, and doesn’t release the domain lock (see section 22.12.2). We can instruct the OCaml native-code compiler of this fact by annotating the external declaration with the attribute [@@noalloc]:

external bar : int -> int -> int = "foo" [@@noalloc]

In this case calling bar from OCaml is as cheap as calling any other OCaml function, except for the fact that the OCaml compiler can’t inline C functions...

22.11.3 Example: calling C library functions without indirection

Using these attributes, it is possible to call C library functions with no indirection. For instance many math functions are defined this way in the OCaml standard library:

external sqrt : float -> float = "caml_sqrt_float" "sqrt"
  [@@unboxed] [@@noalloc]
(** Square root. *)

external exp : float -> float = "caml_exp_float" "exp" [@@unboxed] [@@noalloc]
(** Exponential. *)

external log : float -> float = "caml_log_float" "log" [@@unboxed] [@@noalloc]
(** Natural logarithm. *)

22.12 Advanced topic: multithreading

Using multiple threads (shared-memory concurrency) in a mixed OCaml/C application requires special precautions, which are described in this section.

22.12.1 Registering threads created from C

Callbacks from C to OCaml are possible only if the calling thread is known to the OCaml run-time system. Threads created from OCaml (through the Thread.create function of the system threads library) are automatically known to the run-time system. If the application creates additional threads from C and wishes to callback into OCaml code from these threads, it must first register them with the run-time system. The following functions are declared in the include file <caml/threads.h>.

22.12.2 Parallel execution of long-running C code with systhreads

Domains are the unit of parallelism for OCaml programs. When using the systhreads library, multiple threads might be attached to the same domain. However, at any time, at most one of those thread can be executing OCaml code or C code that uses the OCaml run-time system by domain. Technically, this is enforced by a “domain lock” that any thread must hold while executing such code within a domain.

When OCaml calls the C code implementing a primitive, the domain lock is held, therefore the C code has full access to the facilities of the run-time system. However, no other thread in the same domain can execute OCaml code concurrently with the C code of the primitive. See also chapter 9.6 for the behaviour with multiple domains.

If a C primitive runs for a long time or performs potentially blocking input-output operations, it can explicitly release the domain lock, enabling other OCaml threads in the same domain to run concurrently with its operations. The C code must re-acquire the domain lock before returning to OCaml. This is achieved with the following functions, declared in the include file <caml/threads.h>.

These functions poll for pending signals by calling asynchronous callbacks (section 22.5.3) before releasing and after acquiring the lock. They can therefore execute arbitrary OCaml code including raising an asynchronous exception.

After caml_release_runtime_system() was called and until caml_acquire_runtime_system() is called, the C code must not access any OCaml data, nor call any function of the run-time system, nor call back into OCaml code. Consequently, arguments provided by OCaml to the C primitive must be copied into C data structures before calling caml_release_runtime_system(), and results to be returned to OCaml must be encoded as OCaml values after caml_acquire_runtime_system() returns.

Example: the following C primitive invokes gethostbyname to find the IP address of a host name. The gethostbyname function can block for a long time, so we choose to release the OCaml run-time system while it is running.

CAMLprim stub_gethostbyname(value vname)
{
  CAMLparam1 (vname);
  CAMLlocal1 (vres);
  struct hostent * h;
  char * name;

  /* Copy the string argument to a C string, allocated outside the
     OCaml heap. */
  name = caml_stat_strdup(String_val(vname));
  /* Release the OCaml run-time system */
  caml_release_runtime_system();
  /* Resolve the name */
  h = gethostbyname(name);
  /* Free the copy of the string, which we might as well do before
     acquiring the runtime system to benefit from parallelism. */
  caml_stat_free(name);
  /* Re-acquire the OCaml run-time system */
  caml_acquire_runtime_system();
  /* Encode the relevant fields of h as the OCaml value vres */
  ... /* Omitted */
  /* Return to OCaml */
  CAMLreturn (vres);
}

The macro Caml_state evaluates to the domain state variable, and checks in debug mode that the domain lock is held. Such a check is also placed in normal mode at key entry points of the C API; this is why calling some of the runtime functions and macros without correctly owning the domain lock can result in a fatal error: no domain lock held. The variant Caml_state_opt does not perform any check but evaluates to NULL when the domain lock is not held. This lets you determine whether a thread belonging to a domain currently holds its domain lock, for various purposes.

Callbacks from C to OCaml must be performed while holding the domain lock to the OCaml run-time system. This is naturally the case if the callback is performed by a C primitive that did not release the run-time system. If the C primitive released the run-time system previously, or the callback is performed from other C code that was not invoked from OCaml (e.g. an event loop in a GUI application), the run-time system must be acquired before the callback and released after:

  caml_acquire_runtime_system();
  /* Resolve OCaml function vfun to be invoked */
  /* Build OCaml argument varg to the callback */
  vres = callback(vfun, varg);
  /* Copy relevant parts of result vres to C data structures */
  caml_release_runtime_system();

Note: the acquire and release functions described above were introduced in OCaml 3.12. Older code uses the following historical names, declared in <caml/signals.h>:

Intuition: a “blocking section” is a piece of C code that does not use the OCaml run-time system, typically a blocking input/output operation.

22.13 Advanced topic: interfacing with Windows Unicode APIs

This section contains some general guidelines for writing C stubs that use Windows Unicode APIs.

The OCaml system under Windows can be configured at build time in one of two modes:

In what follows, we say that a string has the OCaml encoding if it is encoded in UTF-8 when in Unicode mode, in the current code page in legacy mode, or is an arbitrary string under Unix. A string has the platform encoding if it is encoded in UTF-16 under Windows or is an arbitrary string under Unix.

From the point of view of the writer of C stubs, the challenges of interacting with Windows Unicode APIs are twofold:

The native C character type under Windows is WCHAR, two bytes wide, while under Unix it is char, one byte wide. A type char_os is defined in <caml/misc.h> that stands for the concrete C character type of each platform. Strings in the platform encoding are of type char_os *.

The following functions are exposed to help write compatible C stubs. To use them, you need to include both <caml/misc.h> and <caml/osdeps.h>.

Note: The strings returned by caml_stat_strdup_to_os and caml_stat_strdup_of_os are allocated using caml_stat_alloc, so they need to be deallocated using caml_stat_free when they are no longer needed.

Example

We want to bind the function getenv in a way that works both under Unix and Windows. Under Unix this function has the prototype:

    char *getenv(const char *);

While the Unicode version under Windows has the prototype:

    WCHAR *_wgetenv(const WCHAR *);

In terms of char_os, both functions take an argument of type char_os * and return a result of the same type. We begin by choosing the right implementation of the function to bind:

#ifdef _WIN32
#define getenv_os _wgetenv
#else
#define getenv_os getenv
#endif

The rest of the binding is the same for both platforms:

#include <caml/mlvalues.h>
#include <caml/misc.h>
#include <caml/alloc.h>
#include <caml/fail.h>
#include <caml/osdeps.h>
#include <stdlib.h>

CAMLprim value stub_getenv(value var_name)
{
  CAMLparam1(var_name);
  CAMLlocal1(var_value);
  char_os *var_name_os, *var_value_os;

  var_name_os = caml_stat_strdup_to_os(String_val(var_name));
  var_value_os = getenv_os(var_name_os);
  caml_stat_free(var_name_os);

  if (var_value_os == NULL)
    caml_raise_not_found();

  var_value = caml_copy_string_of_os(var_value_os);

  CAMLreturn(var_value);
}

22.14 Building mixed C/OCaml libraries: ocamlmklib

The ocamlmklib command facilitates the construction of libraries containing both OCaml code and C code, and usable both in static linking and dynamic linking modes. This command is available under Windows since Objective Caml 3.11 and under other operating systems since Objective Caml 3.03.

The ocamlmklib command takes three kinds of arguments:

It generates the following outputs:

In addition, the following options are recognized:

-cclib, -ccopt, -I, -linkall
These options are passed as is to ocamlc or ocamlopt. See the documentation of these commands.
-rpath, -R, -Wl,-rpath, -Wl,-R
These options are passed as is to the C compiler. Refer to the documentation of the C compiler.
-custom
Force the construction of a statically linked library only, even if dynamic linking is supported.
-failsafe
Fall back to building a statically linked library if a problem occurs while building the shared library (e.g. some of the support libraries are not available as shared libraries).
-Ldir
Add dir to the search path for support libraries (-llib).
-ocamlc cmd
Use cmd instead of ocamlc to call the bytecode compiler.
-ocamlopt cmd
Use cmd instead of ocamlopt to call the native-code compiler.
-o output
Set the name of the generated OCaml library. ocamlmklib will generate output.cma and/or output.cmxa. If not specified, defaults to a.
-oc outputc
Set the name of the generated C library. ocamlmklib will generate liboutputc.so (if shared libraries are supported) and liboutputc.a. If not specified, defaults to the output name given with -o.

On native Windows, the following environment variable is also consulted:

OCAML_FLEXLINK
Alternative executable to use instead of the configured value. Primarily used for bootstrapping.
Example

Consider an OCaml interface to the standard libz C library for reading and writing compressed files. Assume this library resides in /usr/local/zlib. This interface is composed of an OCaml part zip.cmo/zip.cmx and a C part zipstubs.o containing the stub code around the libz entry points. The following command builds the OCaml libraries zip.cma and zip.cmxa, as well as the companion C libraries dllzip.so and libzip.a:

ocamlmklib -o zip zip.cmo zip.cmx zipstubs.o -lz -L/usr/local/zlib

If shared libraries are supported, this performs the following commands:

ocamlc -a -o zip.cma zip.cmo -dllib -lzip \
        -cclib -lzip -cclib -lz -ccopt -L/usr/local/zlib
ocamlopt -a -o zip.cmxa zip.cmx -cclib -lzip \
        -cclib -lzip -cclib -lz -ccopt -L/usr/local/zlib
gcc -shared -o dllzip.so zipstubs.o -lz -L/usr/local/zlib
ar rc libzip.a zipstubs.o

Note: This example is on a Unix system. The exact command lines may be different on other systems.

If shared libraries are not supported, the following commands are performed instead:

ocamlc -a -custom -o zip.cma zip.cmo -cclib -lzip \
        -cclib -lz -ccopt -L/usr/local/zlib
ocamlopt -a -o zip.cmxa zip.cmx -lzip \
        -cclib -lz -ccopt -L/usr/local/zlib
ar rc libzip.a zipstubs.o

Instead of building simultaneously the bytecode library, the native-code library and the C libraries, ocamlmklib can be called three times to build each separately. Thus,

ocamlmklib -o zip zip.cmo -lz -L/usr/local/zlib

builds the bytecode library zip.cma, and

ocamlmklib -o zip zip.cmx -lz -L/usr/local/zlib

builds the native-code library zip.cmxa, and

ocamlmklib -o zip zipstubs.o -lz -L/usr/local/zlib

builds the C libraries dllzip.so and libzip.a. Notice that the support libraries (-lz) and the corresponding options (-L/usr/local/zlib) must be given on all three invocations of ocamlmklib, because they are needed at different times depending on whether shared libraries are supported.

22.15 Cautionary words: the internal runtime API

Not all header available in the caml/ directory were described in previous sections. All those unmentioned headers are part of the internal runtime API, for which there is no stability guarantee. If you really need access to this internal runtime API, this section provides some guidelines that may help you to write code that might not break on every new version of OCaml.

Note

Programmers which come to rely on the internal API for a use-case which they find realistic and useful are encouraged to open a request for improvement on the bug tracker.

22.15.1 Internal variables and CAML_INTERNALS

Since OCaml 4.04, it is possible to get access to every part of the internal runtime API by defining the CAML_INTERNALS macro before loading caml header files. If this macro is not defined, parts of the internal runtime API are hidden.

If you are using internal C variables, do not redefine them by hand. You should import those variables by including the corresponding header files. The representation of those variables has already changed once in OCaml 4.10, and is still under evolution. If your code relies on such internal and brittle properties, it will be broken at some point in time.

For instance, rather than redefining caml_young_limit:

extern int caml_young_limit;

which breaks in OCaml ≄ 4.10, you should include the minor_gc header:

#include <caml/minor_gc.h>

22.15.2 OCaml version macros

Finally, if including the right headers is not enough, or if you need to support version older than OCaml 4.04, the header file caml/version.h should help you to define your own compatibility layer. This file provides few macros defining the current OCaml version. In particular, the OCAML_VERSION macro describes the current version, its format is MmmPP. For example, if you need some specific handling for versions older than 4.10.0, you could write

#include <caml/version.h>
#if OCAML_VERSION >= 41000
...
#else
...
#endif

Chapter 23 Optimisation with Flambda

23.1 Overview

Flambda is the term used to describe a series of optimisation passes provided by the native code compilers as of OCaml 4.03.

Flambda aims to make it easier to write idiomatic OCaml code without incurring performance penalties.

To use the Flambda optimisers it is necessary to pass the -flambda option to the OCaml configure script. (There is no support for a single compiler that can operate in both Flambda and non-Flambda modes.) Code compiled with Flambda cannot be linked into the same program as code compiled without Flambda. Attempting to do this will result in a compiler error.

Whether or not a particular ocamlopt uses Flambda may be determined by invoking it with the -config option and looking for any line starting with “flambda:”. If such a line is present and says “true”, then Flambda is supported, otherwise it is not.

Flambda provides full optimisation across different compilation units, so long as the .cmx files for the dependencies of the unit currently being compiled are available. (A compilation unit corresponds to a single .ml source file.) However it does not yet act entirely as a whole-program compiler: for example, elimination of dead code across a complete set of compilation units is not supported.

Optimisation with Flambda is not currently supported when generating bytecode.

Flambda should not in general affect the semantics of existing programs. Two exceptions to this rule are: possible elimination of pure code that is being benchmarked (see section 23.14) and changes in behaviour of code using unsafe operations (see section 23.15).

Flambda does not yet optimise array or string bounds checks. Neither does it take hints for optimisation from any assertions written by the user in the code.

Consult the Glossary at the end of this chapter for definitions of technical terms used below.

23.2 Command-line flags

The Flambda optimisers provide a variety of command-line flags that may be used to control their behaviour. Detailed descriptions of each flag are given in the referenced sections. Those sections also describe any arguments which the particular flags take.

Commonly-used options:

-O2
Perform more optimisation than usual. Compilation times may be lengthened. (This flag is an abbreviation for a certain set of parameters described in section 23.5.)
-O3
Perform even more optimisation than usual, possibly including unrolling of recursive functions. Compilation times may be significantly lengthened.
-Oclassic
Make inlining decisions at the point of definition of a function rather than at the call site(s). This mirrors the behaviour of OCaml compilers not using Flambda. Compared to compilation using the new Flambda inlining heuristics (for example at -O2) it produces smaller .cmx files, shorter compilation times and code that probably runs rather slower. When using -Oclassic, only the following options described in this section are relevant: -inlining-report and -inline. If any other of the options described in this section are used, the behaviour is undefined and may cause an error in future versions of the compiler.
-inlining-report
Emit .inlining files (one per round of optimisation) showing all of the inliner’s decisions.

Less commonly-used options:

-remove-unused-arguments
Remove unused function arguments even when the argument is not specialised. This may have a small performance penalty. See section 23.10.3.
-unbox-closures
Pass free variables via specialised arguments rather than closures (an optimisation for reducing allocation). See section 23.9.3. This may have a small performance penalty.

Advanced options, only needed for detailed tuning:

-inline
The behaviour depends on whether -Oclassic is used.
-inline-toplevel
The equivalent of -inline but used when speculative inlining starts at toplevel. See section 23.3.6. Not used in -Oclassic mode.
-inline-branch-factor
Controls how the inliner assesses whether a code path is likely to be hot or cold. See section 23.3.5.
-inline-alloc-cost, -inline-branch-cost, -inline-call-cost
Controls how the inliner assesses the runtime performance penalties associated with various operations. See section 23.3.5.
-inline-indirect-cost, -inline-prim-cost
Likewise.
-inline-lifting-benefit
Controls inlining of functors at toplevel. See section 23.3.5.
-inline-max-depth
The maximum depth of any speculative inlining search. See section 23.3.6.
-inline-max-unroll
The maximum depth of any unrolling of recursive functions during any speculative inlining search. See section 23.3.6.
-no-unbox-free-vars-of-closures
Do not unbox closure variables. See section 23.9.1.
-no-unbox-specialised-args
Do not unbox arguments to which functions have been specialised. See section 23.9.2.
-rounds
How many rounds of optimisation to perform. See section 23.2.1.
-unbox-closures-factor
Scaling factor for benefit calculation when using -unbox-closures. See section 23.9.3.
Notes

23.2.1 Specification of optimisation parameters by round

Flambda operates in rounds: one round consists of a certain sequence of transformations that may then be repeated in order to achieve more satisfactory results. The number of rounds can be set manually using the -rounds parameter (although this is not necessary when using predefined optimisation levels such as with -O2 and -O3). For high optimisation the number of rounds might be set at 3 or 4.

Command-line flags that may apply per round, for example those with -cost in the name, accept arguments of the form:

n | round=n[,...]

The flags -Oclassic, -O2 and -O3 are applied before all other flags, meaning that certain parameters may be overridden without having to specify every parameter usually invoked by the given optimisation level.

23.3 Inlining

Inlining refers to the copying of the code of a function to a place where the function is called. The code of the function will be surrounded by bindings of its parameters to the corresponding arguments.

The aims of inlining are:

These goals are often reached not just by inlining itself but also by other optimisations that the compiler is able to perform as a result of inlining.

When a recursive call to a function (within the definition of that function or another in the same mutually-recursive group) is inlined, the procedure is also known as unrolling. This is somewhat akin to loop peeling. For example, given the following code:

let rec fact x =
  if x = 0 then
    1
  else
    x * fact (x - 1)

let n = fact 4

unrolling once at the call site fact 4 produces (with the body of fact unchanged):

let n =
  if 4 = 0 then
    1
  else
    4 * fact (4 - 1)

This simplifies to:

let n = 4 * fact 3

Flambda provides significantly enhanced inlining capabilities relative to previous versions of the compiler.

Aside: when inlining is performed

Inlining is performed together with all of the other Flambda optimisation passes, that is to say, after closure conversion. This has three particular advantages over a potentially more straightforward implementation prior to closure conversion:

23.3.1 Classic inlining heuristic

In -Oclassic mode the behaviour of the Flambda inliner mimics previous versions of the compiler. (Code may still be subject to further optimisations not performed by previous versions of the compiler: functors may be inlined, constants are lifted and unused code is eliminated all as described elsewhere in this chapter. See sections 23.3.3, 23.8.1 and 23.10. At the definition site of a function, the body of the function is measured. It will then be marked as eligible for inlining (and hence inlined at every direct call site) if:

Non-Flambda versions of the compiler cannot inline functions that contain a definition of another function. However -Oclassic does permit this. Further, non-Flambda versions also cannot inline functions that are only themselves exposed as a result of a previous pass of inlining, but again this is permitted by -Oclassic. For example:

module M : sig
  val i : int
end = struct
  let f x =
    let g y = x + y in
    g
  let h = f 3
  let i = h 4  (* h is correctly discovered to be g and inlined *)
end

All of this contrasts with the normal Flambda mode, that is to say without -Oclassic, where:

The Flambda mode is described in the next section.

23.3.2 Overview of “Flambda” inlining heuristics

The Flambda inlining heuristics, used whenever the compiler is configured for Flambda and -Oclassic was not specified, make inlining decisions at call sites. This helps in situations where the context is important. For example:

let f b x =
  if b then
    x
  else
    ... big expression ...

let g x = f true x

In this case, we would like to inline f into g, because a conditional jump can be eliminated and the code size should reduce. If the inlining decision has been made after the declaration of f without seeing the use, its size would have probably made it ineligible for inlining; but at the call site, its final size can be known. Further, this function should probably not be inlined systematically: if b is unknown, or indeed false, there is little benefit to trade off against a large increase in code size. In the existing non-Flambda inliner this isn’t a great problem because chains of inlining were cut off fairly quickly. However it has led to excessive use of overly-large inlining parameters such as -inline 10000.

In more detail, at each call site the following procedure is followed:

Inlining within recursive functions of calls to other functions in the same mutually-recursive group is kept in check by an unrolling depth, described below. This ensures that functions are not unrolled to excess. (Unrolling is only enabled if -O3 optimisation level is selected and/or the -inline-max-unroll flag is passed with an argument greater than zero.)

23.3.3 Handling of specific language constructs

Functors

There is nothing particular about functors that inhibits inlining compared to normal functions. To the inliner, these both look the same, except that functors are marked as such.

Applications of functors at toplevel are biased in favour of inlining. (This bias may be adjusted: see the documentation for -inline-lifting-benefit below.)

Applications of functors not at toplevel, for example in a local module inside some other expression, are treated by the inliner identically to normal function calls.

First-class modules

The inliner will be able to consider inlining a call to a function in a first class module if it knows which particular function is going to be called. The presence of the first-class module record that wraps the set of functions in the module does not per se inhibit inlining.

Objects

Method calls to objects are not at present inlined by Flambda.

23.3.4 Inlining reports

If the -inlining-report option is provided to the compiler then a file will be emitted corresponding to each round of optimisation. For the OCaml source file basename.ml the files are named basename.round.inlining.org, with round a zero-based integer. Inside the files, which are formatted as “org mode”, will be found English prose describing the decisions that the inliner took.

23.3.5 Assessment of inlining benefit

Inlining typically results in an increase in code size, which if left unchecked, may not only lead to grossly large executables and excessive compilation times but also a decrease in performance due to worse locality. As such, the Flambda inliner trades off the change in code size against the expected runtime performance benefit, with the benefit being computed based on the number of operations that the compiler observes may be removed as a result of inlining.

For example given the following code:

let f b x =
  if b then
    x
  else
    ... big expression ...

let g x = f true x

it would be observed that inlining of f would remove:

Formally, an estimate of runtime performance benefit is computed by first summing the cost of the operations that are known to be removed as a result of the inlining and subsequent simplification of the inlined body. The individual costs for the various kinds of operations may be adjusted using the various -inline-...-cost flags as follows. Costs are specified as integers. All of these flags accept a single argument describing such integers using the conventions detailed in section 23.2.1.

-inline-alloc-cost
The cost of an allocation.
-inline-branch-cost
The cost of a branch.
-inline-call-cost
The cost of a direct function call.
-inline-indirect-cost
The cost of an indirect function call.
-inline-prim-cost
The cost of a primitive. Primitives encompass operations including arithmetic and memory access.

(Default values are described in section 23.5 below.)

The initial benefit value is then scaled by a factor that attempts to compensate for the fact that the current point in the code, if under some number of conditional branches, may be cold. (Flambda does not currently compute hot and cold paths.) The factor—the estimated probability that the inliner really is on a hot path—is calculated as 1/(1 + f)d, where f is set by -inline-branch-factor and d is the nesting depth of branches at the current point. As the inliner descends into more deeply-nested branches, the benefit of inlining thus lessens.

The resulting benefit value is known as the estimated benefit.

The change in code size is also estimated: morally speaking it should be the change in machine code size, but since that is not available to the inliner, an approximation is used.

If the estimated benefit exceeds the increase in code size then the inlined version of the function will be kept. Otherwise the function will not be inlined.

Applications of functors at toplevel will be given an additional benefit (which may be controlled by the -inline-lifting-benefit flag) to bias inlining in such situations towards keeping the inlined version.

23.3.6 Control of speculation

As described above, there are three parameters that restrict the search for inlining opportunities during speculation:

These parameters are ultimately bounded by the arguments provided to the corresponding command-line flags (or their default values):

Note in particular that -inline does not have the meaning that it has in the previous compiler or in -Oclassic mode. In both of those situations -inline was effectively some kind of basic assessment of inlining benefit. However in Flambda inlining mode it corresponds to a constraint on the search; the assessment of benefit is independent, as described above.

When speculation starts the inlining threshold starts at the value set by -inline (or -inline-toplevel if appropriate, see above). Upon making a speculative inlining decision the threshold is reduced by the code size of the function being inlined. If the threshold becomes exhausted, at or below zero, no further speculation will be performed.

The inlining depth starts at zero and is increased by one every time the inliner descends into another function. It is then decreased by one every time the inliner leaves such function. If the depth exceeds the value set by -inline-max-depth then speculation stops. This parameter is intended as a general backstop for situations where the inlining threshold does not control the search sufficiently.

The unrolling depth applies to calls within the same mutually-recursive group of functions. Each time an inlining of such a call is performed the depth is incremented by one when examining the resulting body. If the depth reaches the limit set by -inline-max-unroll then speculation stops.

23.4 Specialisation

The inliner may discover a call site to a recursive function where something is known about the arguments: for example, they may be equal to some other variables currently in scope. In this situation it may be beneficial to specialise the function to those arguments. This is done by copying the declaration of the function (and any others involved in any same mutually-recursive declaration) and noting the extra information about the arguments. The arguments augmented by this information are known as specialised arguments. In order to try to ensure that specialisation is not performed uselessly, arguments are only specialised if it can be shown that they are invariant: in other words, during the execution of the recursive function(s) themselves, the arguments never change.

Unless overridden by an attribute (see below), specialisation of a function will not be attempted if:

The compiler can prove invariance of function arguments across multiple functions within a recursive group (although this has some limitations, as shown by the example below).

It should be noted that the unboxing of closures pass (see below) can introduce specialised arguments on non-recursive functions. (No other place in the compiler currently does this.)

Example: the well-known List.iter function

This function might be written like so:

let rec iter f l =
  match l with
  | [] -> ()
  | h :: t ->
    f h;
    iter f t

and used like this:

let print_int x =
  print_endline (Int.to_string x)

let run xs =
  iter print_int (List.rev xs)

The argument f to iter is invariant so the function may be specialised:

let run xs =
  let rec iter' f l =
    (* The compiler knows: f holds the same value as foo throughout iter'. *)
    match l with
    | [] -> ()
    | h :: t ->
      f h;
      iter' f t
  in
  iter' print_int (List.rev xs)

The compiler notes down that for the function iter’, the argument f is specialised to the constant closure print_int. This means that the body of iter’ may be simplified:

let run xs =
  let rec iter' f l =
    (* The compiler knows: f holds the same value as foo throughout iter'. *)
    match l with
    | [] -> ()
    | h :: t ->
      print_int h;  (* this is now a direct call *)
      iter' f t
  in
  iter' print_int (List.rev xs)

The call to print_int can indeed be inlined:

let run xs =
  let rec iter' f l =
    (* The compiler knows: f holds the same value as foo throughout iter'. *)
    match l with
    | [] -> ()
    | h :: t ->
      print_endline (Int.to_string h);
      iter' f t
  in
  iter' print_int (List.rev xs)

The unused specialised argument f may now be removed, leaving:

let run xs =
  let rec iter' l =
    match l with
    | [] -> ()
    | h :: t ->
      print_endline (Int.to_string h);
      iter' t
  in
  iter' (List.rev xs)
Aside on invariant parameters.

The compiler cannot currently detect invariance in cases such as the following.

let rec iter_swap f g l =
  match l with
  | [] -> ()
  | 0 :: t ->
    iter_swap g f l
  | h :: t ->
    f h;
    iter_swap f g t

23.4.1 Assessment of specialisation benefit

The benefit of specialisation is assessed in a similar way as for inlining. Specialised argument information may mean that the body of the function being specialised can be simplified: the removed operations are accumulated into a benefit. This, together with the size of the duplicated (specialised) function declaration, is then assessed against the size of the call to the original function.

23.5 Default settings of parameters

The default settings (when not using -Oclassic) are for one round of optimisation using the following parameters.

ParameterSetting
-inline10
-inline-branch-factor0.1
-inline-alloc-cost7
-inline-branch-cost5
-inline-call-cost5
-inline-indirect-cost4
-inline-prim-cost3
-inline-lifting-benefit1300
-inline-toplevel160
-inline-max-depth1
-inline-max-unroll0
-unbox-closures-factor10

23.5.1 Settings at -O2 optimisation level

When -O2 is specified two rounds of optimisation are performed. The first round uses the default parameters (see above). The second uses the following parameters.

ParameterSetting
-inline25
-inline-branch-factorSame as default
-inline-alloc-costDouble the default
-inline-branch-costDouble the default
-inline-call-costDouble the default
-inline-indirect-costDouble the default
-inline-prim-costDouble the default
-inline-lifting-benefitSame as default
-inline-toplevel400
-inline-max-depth2
-inline-max-unrollSame as default
-unbox-closures-factorSame as default

23.5.2 Settings at -O3 optimisation level

When -O3 is specified three rounds of optimisation are performed. The first two rounds are as for -O2. The third round uses the following parameters.

ParameterSetting
-inline50
-inline-branch-factorSame as default
-inline-alloc-costTriple the default
-inline-branch-costTriple the default
-inline-call-costTriple the default
-inline-indirect-costTriple the default
-inline-prim-costTriple the default
-inline-lifting-benefitSame as default
-inline-toplevel800
-inline-max-depth3
-inline-max-unroll1
-unbox-closures-factorSame as default

23.6 Manual control of inlining and specialisation

Should the inliner prove recalcitrant and refuse to inline a particular function, or if the observed inlining decisions are not to the programmer’s satisfaction for some other reason, inlining behaviour can be dictated by the programmer directly in the source code. One example where this might be appropriate is when the programmer, but not the compiler, knows that a particular function call is on a cold code path. It might be desirable to prevent inlining of the function so that the code size along the hot path is kept smaller, so as to increase locality.

The inliner is directed using attributes. For non-recursive functions (and one-step unrolling of recursive functions, although @unroll is more clear for this purpose) the following are supported:

@@inline always or @@inline never
Attached to a declaration of a function or functor, these direct the inliner to either always or never inline, irrespective of the size/benefit calculation. (If the function is recursive then the body is substituted and no special action is taken for the recursive call site(s).) @@inline with no argument is equivalent to @@inline always.
@inlined always or @inlined never
Attached to a function application, these direct the inliner likewise. These attributes at call sites override any other attribute that may be present on the corresponding declaration. @inlined with no argument is equivalent to @inlined always. @@inlined hint is equivalent to @@inline always except that it will not trigger warning 55 if the function application cannot be inlined.

For recursive functions the relevant attributes are:

@@specialise always or @@specialise never
Attached to a declaration of a function or functor, this directs the inliner to either always or never specialise the function so long as it has appropriate contextual knowledge, irrespective of the size/benefit calculation. @@specialise with no argument is equivalent to @@specialise always.
@specialised always or @specialised never
Attached to a function application, this directs the inliner likewise. This attribute at a call site overrides any other attribute that may be present on the corresponding declaration. (Note that the function will still only be specialised if there exist one or more invariant parameters whose values are known.) @specialised with no argument is equivalent to @specialised always.
@unrolled n
This attribute is attached to a function application and always takes an integer argument. Each time the inliner sees the attribute it behaves as follows: As such, n behaves as the “maximum depth of unrolling”.

A compiler warning will be emitted if it was found impossible to obey an annotation from an @inlined or @specialised attribute.

Example showing correct placement of attributes
module F (M : sig type t end) = struct
  let[@inline never] bar x =
    x * 3

  let foo x =
    (bar [@inlined]) (42 + x)
end [@@inline never]

module X = F [@inlined] (struct type t = int end)

23.7 Simplification

Simplification, which is run in conjunction with inlining, propagates information (known as approximations) about which variables hold what values at runtime. Certain relationships between variables and symbols are also tracked: for example, some variable may be known to always hold the same value as some other variable; or perhaps some variable may be known to always hold the value pointed to by some symbol.

The propagation can help to eliminate allocations in cases such as:

let f x y =
  ...
  let p = x, y in
  ...
  ... (fst p) ... (snd p) ...

The projections from p may be replaced by uses of the variables x and y, potentially meaning that p becomes unused.

The propagation performed by the simplification pass is also important for discovering which functions flow to indirect call sites. This can enable the transformation of such call sites into direct call sites, which makes them eligible for an inlining transformation.

Note that no information is propagated about the contents of strings, even in safe-string mode, because it cannot yet be guaranteed that they are immutable throughout a given program.

23.8 Other code motion transformations

23.8.1 Lifting of constants

Expressions found to be constant will be lifted to symbol bindings—that is to say, they will be statically allocated in the object file—when they evaluate to boxed values. Such constants may be straightforward numeric constants, such as the floating-point number 42.0, or more complicated values such as constant closures.

Lifting of constants to toplevel reduces allocation at runtime.

The compiler aims to share constants lifted to toplevel such that there are no duplicate definitions. However if .cmx files are hidden from the compiler then maximal sharing may not be possible.

Notes about float arrays

The following language semantics apply specifically to constant float arrays. (By “constant float array” is meant an array consisting entirely of floating point numbers that are known at compile time. A common case is a literal such as [| 42.0; 43.0; |].

23.8.2 Lifting of toplevel let bindings

Toplevel let-expressions may be lifted to symbol bindings to ensure that the corresponding bound variables are not captured by closures. If the defining expression of a given binding is found to be constant, it is bound as such (the technical term is a let-symbol binding).

Otherwise, the symbol is bound to a (statically-allocated) preallocated block containing one field. At runtime, the defining expression will be evaluated and the first field of the block filled with the resulting value. This initialise-symbol binding causes one extra indirection but ensures, by virtue of the symbol’s address being known at compile time, that uses of the value are not captured by closures.

It should be noted that the blocks corresponding to initialise-symbol bindings are kept alive forever, by virtue of them occurring in a static table of GC roots within the object file. This extended lifetime of expressions may on occasion be surprising. If it is desired to create some non-constant value (for example when writing GC tests) that does not have this extended lifetime, then it may be created and used inside a function, with the application point of that function (perhaps at toplevel)—or indeed the function declaration itself—marked as to never be inlined. This technique prevents lifting of the definition of the value in question (assuming of course that it is not constant).

23.9 Unboxing transformations

The transformations in this section relate to the splitting apart of boxed (that is to say, non-immediate) values. They are largely intended to reduce allocation, which tends to result in a runtime performance profile with lower variance and smaller tails.

23.9.1 Unboxing of closure variables

This transformation is enabled unless -no-unbox-free-vars-of-closures is provided.

Variables that appear in closure environments may themselves be boxed values. As such, they may be split into further closure variables, each of which corresponds to some projection from the original closure variable(s). This transformation is called unboxing of closure variables or unboxing of free variables of closures. It is only applied when there is reasonable certainty that there are no uses of the boxed free variable itself within the corresponding function bodies.

Example:

In the following code, the compiler observes that the closure returned from the function f contains a variable pair (free in the body of f) that may be split into two separate variables.

let f x0 x1 =
  let pair = x0, x1 in
  Printf.printf "foo\n";
  fun y ->
    fst pair + snd pair + y

After some simplification one obtains:

let f x0 x1 =
  let pair_0 = x0 in
  let pair_1 = x1 in
  Printf.printf "foo\n";
  fun y ->
    pair_0 + pair_1 + y

and then:

let f x0 x1 =
  Printf.printf "foo\n";
  fun y ->
    x0 + x1 + y

The allocation of the pair has been eliminated.

This transformation does not operate if it would cause the closure to contain more than twice as many closure variables as it did beforehand.

23.9.2 Unboxing of specialised arguments

This transformation is enabled unless -no-unbox-specialised-args is provided.

It may become the case during compilation that one or more invariant arguments to a function become specialised to a particular value. When such values are themselves boxed the corresponding specialised arguments may be split into more specialised arguments corresponding to the projections out of the boxed value that occur within the function body. This transformation is called unboxing of specialised arguments. It is only applied when there is reasonable certainty that the boxed argument itself is unused within the function.

If the function in question is involved in a recursive group then unboxing of specialised arguments may be immediately replicated across the group based on the dataflow between invariant arguments.

Example:

Having been given the following code, the compiler will inline loop into f, and then observe inv being invariant and always the pair formed by adding 42 and 43 to the argument x of the function f.

let rec loop inv xs =
  match xs with
  | [] -> fst inv + snd inv
  | x::xs -> x + loop2 xs inv
and loop2 ys inv =
  match ys with
  | [] -> 4
  | y::ys -> y - loop inv ys

let f x =
  Printf.printf "%d\n" (loop (x + 42, x + 43) [1; 2; 3])

Since the functions have sufficiently few arguments, more specialised arguments will be added. After some simplification one obtains:

let f x =
  let rec loop' xs inv_0 inv_1 =
    match xs with
    | [] -> inv_0 + inv_1
    | x::xs -> x + loop2' xs inv_0 inv_1
  and loop2' ys inv_0 inv_1 =
    match ys with
    | [] -> 4
    | y::ys -> y - loop' ys inv_0 inv_1
  in
  Printf.printf "%d\n" (loop' [1; 2; 3] (x + 42) (x + 43))

The allocation of the pair within f has been removed. (Since the two closures for loop’ and loop2’ are constant they will also be lifted to toplevel with no runtime allocation penalty. This would also happen without having run the transformation to unbox specialise arguments.)

The transformation to unbox specialised arguments never introduces extra allocation.

The transformation will not unbox arguments if it would result in the original function having sufficiently many arguments so as to inhibit tail-call optimisation.

The transformation is implemented by creating a wrapper function that accepts the original arguments. Meanwhile, the original function is renamed and extra arguments are added corresponding to the unboxed specialised arguments; this new function is called from the wrapper. The wrapper will then be inlined at direct call sites. Indeed, all call sites will be direct unless -unbox-closures is being used, since they will have been generated by the compiler when originally specialising the function. (In the case of -unbox-closures other functions may appear with specialised arguments; in this case there may be indirect calls and these will incur a small penalty owing to having to bounce through the wrapper. The technique of direct call surrogates used for -unbox-closures is not used by the transformation to unbox specialised arguments.)

23.9.3 Unboxing of closures

This transformation is not enabled by default. It may be enabled using the -unbox-closures flag.

The transformation replaces closure variables by specialised arguments. The aim is to cause more closures to become closed. It is particularly applicable, as a means of reducing allocation, where the function concerned cannot be inlined or specialised. For example, some non-recursive function might be too large to inline; or some recursive function might offer no opportunities for specialisation perhaps because its only argument is one of type unit.

At present there may be a small penalty in terms of actual runtime performance when this transformation is enabled, although more stable performance may be obtained due to reduced allocation. It is recommended that developers experiment to determine whether the option is beneficial for their code. (It is expected that in the future it will be possible for the performance degradation to be removed.)

Simple example:

In the following code (which might typically occur when g is too large to inline) the value of x would usually be communicated to the application of the + function via the closure of g.

let f x =
  let g y =
    x + y
  in
  (g [@inlined never]) 42

Unboxing of the closure causes the value for x inside g to be passed as an argument to g rather than through its closure. This means that the closure of g becomes constant and may be lifted to toplevel, eliminating the runtime allocation.

The transformation is implemented by adding a new wrapper function in the manner of that used when unboxing specialised arguments. The closure variables are still free in the wrapper, but the intention is that when the wrapper is inlined at direct call sites, the relevant values are passed directly to the main function via the new specialised arguments.

Adding such a wrapper will penalise indirect calls to the function (which might exist in arbitrary places; remember that this transformation is not for example applied only on functions the compiler has produced as a result of specialisation) since such calls will bounce through the wrapper. To mitigate this, if a function is small enough when weighed up against the number of free variables being removed, it will be duplicated by the transformation to obtain two versions: the original (used for indirect calls, since we can do no better) and the wrapper/rewritten function pair as described in the previous paragraph. The wrapper/rewritten function pair will only be used at direct call sites of the function. (The wrapper in this case is known as a direct call surrogate, since it takes the place of another function—the unchanged version used for indirect calls—at direct call sites.)

The -unbox-closures-factor command line flag, which takes an integer, may be used to adjust the point at which a function is deemed large enough to be ineligible for duplication. The benefit of duplication is scaled by the integer before being evaluated against the size.

Harder example:

In the following code, there are two closure variables that would typically cause closure allocations. One is called fv and occurs inside the function baz; the other is called z and occurs inside the function bar. In this toy (yet sophisticated) example we again use an attribute to simulate the typical situation where the first argument of baz is too large to inline.

let foo c =
  let rec bar zs fv =
    match zs with
    | [] -> []
    | z::zs ->
      let rec baz f = function
        | [] -> []
        | a::l -> let r = fv + ((f [@inlined never]) a) in r :: baz f l
      in
      (map2 (fun y -> z + y) [z; 2; 3; 4]) @ bar zs fv
  in
  Printf.printf "%d" (List.length (bar [1; 2; 3; 4] c))

The code resulting from applying -O3 -unbox-closures to this code passes the free variables via function arguments in order to eliminate all closure allocation in this example (aside from any that might be performed inside printf).

23.10 Removal of unused code and values

23.10.1 Removal of redundant let expressions

The simplification pass removes unused let bindings so long as their corresponding defining expressions have “no effects”. See the section “Treatment of effects” below for the precise definition of this term.

23.10.2 Removal of redundant program constructs

This transformation is analogous to the removal of let-expressions whose defining expressions have no effects. It operates instead on symbol bindings, removing those that have no effects.

23.10.3 Removal of unused arguments

This transformation is only enabled by default for specialised arguments. It may be enabled for all arguments using the -remove-unused-arguments flag.

The pass analyses functions to determine which arguments are unused. Removal is effected by creating a wrapper function, which will be inlined at every direct call site, that accepts the original arguments and then discards the unused ones before calling the original function. As a consequence, this transformation may be detrimental if the original function is usually indirectly called, since such calls will now bounce through the wrapper. (The technique of direct call surrogates used to reduce this penalty during unboxing of closure variables (see above) does not yet apply to the pass that removes unused arguments.)

23.10.4 Removal of unused closure variables

This transformation performs an analysis across the whole compilation unit to determine whether there exist closure variables that are never used. Such closure variables are then eliminated. (Note that this has to be a whole-unit analysis because a projection of a closure variable from some particular closure may have propagated to an arbitrary location within the code due to inlining.)

23.11 Other code transformations

23.11.1 Transformation of non-escaping references into mutable variables

Flambda performs a simple analysis analogous to that performed elsewhere in the compiler that can transform refs into mutable variables that may then be held in registers (or on the stack as appropriate) rather than being allocated on the OCaml heap. This only happens so long as the reference concerned can be shown to not escape from its defining scope.

23.11.2 Substitution of closure variables for specialised arguments

This transformation discovers closure variables that are known to be equal to specialised arguments. Such closure variables are replaced by the specialised arguments; the closure variables may then be removed by the “removal of unused closure variables” pass (see below).

23.12 Treatment of effects

The Flambda optimisers classify expressions in order to determine whether an expression:

This is done by forming judgements on the effects and the coeffects that might be performed were the expression to be executed. Effects talk about how the expression might affect the world; coeffects talk about how the world might affect the expression.

Effects are classified as follows:

No effects:
The expression does not change the observable state of the world. For example, it must not write to any mutable storage, call arbitrary external functions or change control flow (e.g. by raising an exception). Note that allocation is not classed as having “no effects” (see below).
Only generative effects:
The expression does not change the observable state of the world save for possibly affecting the state of the garbage collector by performing an allocation. Expressions that only have generative effects and whose results are unused may be eliminated by the compiler. However, unlike expressions with “no effects”, such expressions will never be eligible for duplication.
Arbitrary effects:
All other expressions.

There is a single classification for coeffects:

No coeffects:
The expression does not observe the effects (in the sense described above) of other expressions. For example, it must not read from any mutable storage or call arbitrary external functions.

It is assumed in the compiler that, subject to data dependencies, expressions with neither effects nor coeffects may be reordered with respect to other expressions.

23.13 Compilation of statically-allocated modules

Compilation of modules that are able to be statically allocated (for example, the module corresponding to an entire compilation unit, as opposed to a first class module dependent on values computed at runtime) initially follows the strategy used for bytecode. A sequence of let-bindings, which may be interspersed with arbitrary effects, surrounds a record creation that becomes the module block. The Flambda-specific transformation follows: these bindings are lifted to toplevel symbols, as described above.

23.14 Inhibition of optimisation

Especially when writing benchmarking suites that run non-side-effecting algorithms in loops, it may be found that the optimiser entirely elides the code being benchmarked. This behaviour can be prevented by using the Sys.opaque_identity function (which indeed behaves as a normal OCaml function and does not possess any “magic” semantics). The documentation of the Sys module should be consulted for further details.

23.15 Use of unsafe operations

The behaviour of the Flambda simplification pass means that certain unsafe operations, which may without Flambda or when using previous versions of the compiler be safe, must not be used. This specifically refers to functions found in the Obj module.

In particular, it is forbidden to change any value (for example using Obj.set_field or Obj.set_tag) that is not mutable. (Values returned from C stubs are always treated as mutable.) The compiler will emit warning 59 if it detects such a write—but it cannot warn in all cases. Here is an example of code that will trigger the warning:

let f x =
  let a = 42, x in
  (Obj.magic a : int ref) := 1;
  fst a

The reason this is unsafe is because the simplification pass believes that fst a holds the value 42; and indeed it must, unless type soundness has been broken via unsafe operations.

If it must be the case that code has to be written that triggers warning 59, but the code is known to actually be correct (for some definition of correct), then Sys.opaque_identity may be used to wrap the value before unsafe operations are performed upon it. Great care must be taken when doing this to ensure that the opacity is added at the correct place. It must be emphasised that this use of Sys.opaque_identity is only for exceptional cases. It should not be used in normal code or to try to guide the optimiser.

As an example, this code will return the integer 1:

let f x =
  let a = Sys.opaque_identity (42, x) in
  (Obj.magic a : int ref) := 1;
  fst a

However the following code will still return 42:

let f x =
  let a = 42, x in
  Sys.opaque_identity (Obj.magic a : int ref) := 1;
  fst a

High levels of inlining performed by Flambda may expose bugs in code thought previously to be correct. Take care, for example, not to add type annotations that claim some mutable value is always immediate if it might be possible for an unsafe operation to update it to a boxed value.

23.16 Glossary

The following terminology is used in this chapter of the manual.

Call site
See direct call site and indirect call site below.
Closed function
A function whose body has no free variables except its parameters and any to which are bound other functions within the same (possibly mutually-recursive) declaration.
Closure
The runtime representation of a function. This includes pointers to the code of the function together with the values of any variables that are used in the body of the function but actually defined outside of the function, in the enclosing scope. The values of such variables, collectively known as the environment, are required because the function may be invoked from a place where the original bindings of such variables are no longer in scope. A group of possibly mutually-recursive functions defined using let rec all share a single closure. (Note to developers: in the Flambda source code a closure always corresponds to a single function; a set of closures refers to a group of such.)
Closure variable
A member of the environment held within the closure of a given function.
Constant
Some entity (typically an expression) the value of which is known by the compiler at compile time. Constantness may be explicit from the source code or inferred by the Flambda optimisers.
Constant closure
A closure that is statically allocated in an object file. It is almost always the case that the environment portion of such a closure is empty.
Defining expression
The expression e in let x = e in e’.
Direct call site
A place in a program’s code where a function is called and it is known at compile time which function it will always be.
Indirect call site
A place in a program’s code where a function is called but is not known to be a direct call site.
Program
A collection of symbol bindings forming the definition of a single compilation unit (i.e. .cmx file).
Specialised argument
An argument to a function that is known to always hold a particular value at runtime. These are introduced by the inliner when specialising recursive functions; and the unbox-closures pass. (See section 23.4.)
Symbol
A name referencing a particular place in an object file or executable image. At that particular place will be some constant value. Symbols may be examined using operating system-specific tools (for example objdump on Linux).
Symbol binding
Analogous to a let-expression but working at the level of symbols defined in the object file. The address of a symbol is fixed, but it may be bound to both constant and non-constant expressions.
Toplevel
An expression in the current program which is not enclosed within any function declaration.
Variable
A named entity to which some OCaml value is bound by a let expression, pattern-matching construction, or similar.

Chapter 24 Fuzzing with afl-fuzz

24.1 Overview

American fuzzy lop (“afl-fuzz”) is a fuzzer, a tool for testing software by providing randomly-generated inputs, searching for those inputs which cause the program to crash.

Unlike most fuzzers, afl-fuzz observes the internal behaviour of the program being tested, and adjusts the test cases it generates to trigger unexplored execution paths. As a result, test cases generated by afl-fuzz cover more of the possible behaviours of the tested program than other fuzzers.

This requires that programs to be tested are instrumented to communicate with afl-fuzz. The native-code compiler “ocamlopt” can generate such instrumentation, allowing afl-fuzz to be used against programs written in OCaml.

For more information on afl-fuzz, see the website at http://lcamtuf.coredump.cx/afl/.

24.2 Generating instrumentation

The instrumentation that afl-fuzz requires is not generated by default, and must be explicitly enabled, by passing the -afl-instrument option to ocamlopt.

To fuzz a large system without modifying build tools, OCaml’s configure script also accepts the afl-instrument option. If OCaml is configured with afl-instrument, then all programs compiled by ocamlopt will be instrumented.

24.2.1 Advanced options

In rare cases, it is useful to control the amount of instrumentation generated. By passing the -afl-inst-ratio N argument to ocamlopt with N less than 100, instrumentation can be generated for only N% of branches. (See the afl-fuzz documentation on the parameter AFL_INST_RATIO for the precise effect of this).

24.3 Example

As an example, we fuzz-test the following program, readline.ml:

let _ =
  let s = read_line () in
  match Array.to_list (Array.init (String.length s) (String.get s)) with
    ['s'; 'e'; 'c'; 'r'; 'e'; 't'; ' '; 'c'; 'o'; 'd'; 'e'] -> failwith "uh oh"
  | _ -> ()

There is a single input (the string “secret code”) which causes this program to crash, but finding it by blind random search is infeasible.

Instead, we compile with afl-fuzz instrumentation enabled:

ocamlopt -afl-instrument readline.ml -o readline

Next, we run the program under afl-fuzz:

mkdir input
echo asdf > input/testcase
mkdir output
afl-fuzz -m none -i input -o output ./readline

By inspecting instrumentation output, the fuzzer finds the crashing input quickly.

Note: To fuzz-test an OCaml program with afl-fuzz, passing the option -m none is required to disable afl-fuzz’s default 50MB virtual memory limit.

Chapter 25 Runtime tracing with runtime events

This chapter describes the runtime events tracing system which enables continuous extraction of performance information from the OCaml runtime with very low overhead. The system and interfaces are low-level and tightly coupled to the runtime implementation, it is intended for end-users to rely on tooling to consume and visualise data of interest.

Data emitted includes:

25.1 Overview

There are three main classes of events emitted by the runtime events system:

Spans
Events spanning over a duration in time. For example, the runtime events tracing system emits a span event that starts when a minor collection begins in the OCaml garbage collector and ends when the collection is completed. Spans can contain other spans, e.g other span events may be emitted that begin after a minor collection has begun and end before it does.
Lifecycle events
Events that occur at a moment in time. For example, when a domain terminates, a corresponding lifecycle event is emitted.
Counters
Events that include a measurement of some quantity of interest. For example, the number of words promoted from the minor to the major heap during the last minor garbage collection is emitted as a counter event.

The runtime events tracing system is designed to be used in different contexts:

Self monitoring
OCaml programs and libraries can install their own callbacks to listen for runtime events and react to them programmatically, for example, to export events to disk or over the network.
External monitoring
An external process can consume the runtime events of an OCaml program whose runtime tracing system has been enabled by setting the corresponding environment variable.

The runtime events tracing system logs events to a ring buffer. Consequently, old events are being overwritten by new events. Consumers can either continuously consume events or choose to only do so in response to some circumstance, e.g if a particular query or operation takes longer than expected to complete.

25.2 Architecture

The runtime tracing system conceptually consists of two parts: 1) the probes which emit events and 2) the events transport that ingests and transports these events.

25.2.1 Probes

Probes collect events from the runtime system. These are further split in to two sets: 1) probes that are always available and 2) probes that are only available in the instrumented runtime. Probes in the instrumented runtime are primarily of interest to developers of the OCaml runtime and garbage collector and, at present, only consist of major heap allocation size counter events.

The full set of events emitted by probes and their documentation can be found in Module Runtime_events.

25.2.2 Events transport

The events transport part of the system ingests events emitted by the probes and makes them available to consumers.

Ring buffers

Events are transported using a data structure known as a ring buffer. This data structure consists of two pointers into a linear backing array, the tail pointer points to a location where new events can be written and the head pointer points to the oldest event in the buffer that can be read. When insufficient space is available in the backing array to write new events, the head pointer is advanced and the oldest events are overwritten by new ones.

The ring buffer implementation used in runtime events can be written by at most one producer at a time but can be read simultaneously by multiple consumers without coordination from the producer. There is a unique ring buffer for every running domain and, on domain termination, ring buffers may be re-used for newly spawned domains. The ring buffers themselves are stored in a memory-mapped file with the processes identifier as the name and the extension .events, this enables them to be read from outside the main OCaml process. See Runtime_events for more information.

Consumption APIs

The runtime event tracing system provides both OCaml and C APIs which are cursor-based and polling-driven. The high-level process for consuming events is as follows:

  1. A cursor is created via Runtime_events.create_cursor for either the current process or an external process (specified by a path and PID).
  2. Runtime_events.Callbacks.create is called to register a callback function to receive the events.
  3. The cursor is polled via Runtime_events.read_poll using the callbacks created in the previous step. For each matching event in the ring buffers, the provided callback functions are called.

25.3 Usage

25.3.1 With OCaml APIs

We start with a simple example that prints the name, begin and end times of events emitted by the runtime event tracing system:

let runtime_begin _ ts phase =
    Printf.printf "Begin\t%s\t%Ld\n"
        (Runtime_events.runtime_phase_name phase)
        (Runtime_events.Timestamp.to_int64 ts)

let runtime_end _ ts phase =
    Printf.printf "End\t%s\t%Ld\n"
        (Runtime_events.runtime_phase_name phase)
        (Runtime_events.Timestamp.to_int64 ts)

let () =
    Runtime_events.start ();
    let cursor = Runtime_events.create_cursor None in
    let callbacks = Runtime_events.Callbacks.create ~runtime_begin ~runtime_end ()
    in
    while true do
        let list_ref = ref [] in (* for later fake GC work *)
        for _ = 1 to 100 do
            (* here we do some fake GC work *)
            list_ref := [];
            for _ = 1 to 10 do
                list_ref := (Sys.opaque_identity(ref 42)) :: !list_ref
            done;
            Gc.full_major ();
        done;
        ignore(Runtime_events.read_poll cursor callbacks None);
        Unix.sleep 1
    done

The next step is to compile and link the program with the runtime_events library. This can be done as follows:

       ocamlopt -I +runtime_events -I +unix unix.cmxa runtime_events.cmxa
            example.ml -o example

When using the dune build system, this example can be built as follows:

(executable
 (name example)
 (modules example)
 (libraries unix runtime_events))

Running the compiled binary of the example gives an output similar to:

Begin   explicit_gc_full_major  24086187297852
Begin   stw_leader      24086187298594
Begin   minor   24086187299404
Begin   minor_global_roots      24086187299807
End     minor_global_roots      24086187331461
Begin   minor_remembered_set    24086187331631
Begin   minor_finalizers_oldify 24086187544312
End     minor_finalizers_oldify 24086187544704
Begin   minor_remembered_set_promote    24086187544879
End     minor_remembered_set_promote    24086187606414
End     minor_remembered_set    24086187606584
Begin   minor_finalizers_admin  24086187606854
End     minor_finalizers_admin  24086187607152
Begin   minor_local_roots       24086187607329
Begin   minor_local_roots_promote       24086187609699
End     minor_local_roots_promote       24086187610539
End     minor_local_roots       24086187610709
End     minor   24086187611746
Begin   minor_clear     24086187612238
End     minor_clear     24086187612580
End     stw_leader      24086187613209
...

This is an example of self-monitoring, where a program explicitly starts listening to runtime events and monitors itself.

For external monitoring, a program does not need to be aware of the existence of runtime events. Runtime events can be controlled via the environment variable OCAML_RUNTIME_EVENTS_START which, when set, will cause the runtime tracing system to be started at program initialization.

We could remove Runtime_events.start (); from the previous example and, instead, call the program as below to produce the same result:

OCAML_RUNTIME_EVENTS_START=1 ./example

Environment variables

Environment variables can be used to control different aspects of the runtime event tracing system. The following environment variables are available:

The size of the runtime events ring buffers can be configured via OCAMLRUNPARAM, see section 15.2 for more information.

Building with the instrumented runtime

To receive events that are only available in the instrumented runtime, the OCaml program needs to be compiled and linked against the instrumented runtime. For our example program from earlier, this is achieved as follows:

ocamlopt -runtime-variant i -I +runtime_events -I +unix unix.cmxa runtime_events.cmxa example.ml -o example

And for dune:

(executable
 (name example)
 (modules example)
 (flags "-runtime-variant=i")
 (libraries unix runtime_events))

25.3.2 With tooling

Programmatic access to events is intended primarily for writers of observability libraries and tooling that end-users use. The flexible API enables use of the performance data from runtime events for logging and monitoring purposes.

In this section we cover several utilities in the runtime_events_tools package which provide simple ways of extracting and summarising data from runtime events. The trace utility in particular produces similar data to the previous ’eventlog’ instrumentation system available in OCaml 4.12 to 4.14.

First, install runtime_events_tools in an OCaml 5.0+ opam switch:

opam install runtime_events_tools

This should install the olly tool in your path. You can now generate runtime traces for programs compiled with OCaml 5.0+ using the trace subcommand:

olly trace trace.json 'your_program.exe .. args ..'

Runtime tracing data will be generated in the json Trace Event Format to trace.json. This can then be loaded into the Chrome tracing viewer or into Perfetto to visualize the collected trace.

Measuring GC latency

The olly utility also includes a latency subcommand which consumes runtime events data and on program completion emits a parseable histogram summary of pause durations. It can be run as follows:

olly latency 'your_program.exe .. args ..'

This should produce an output similar to the following:

GC latency profile:
#[Mean (ms):	2.46,	 Stddev (ms):	3.87]
#[Min (ms):	0.01,	 max (ms):	9.17]

Percentile 	 Latency (ms)
25.0000 	 0.01
50.0000 	 0.23
60.0000 	 0.23
70.0000 	 0.45
75.0000 	 0.45
80.0000 	 0.45
85.0000 	 0.45
90.0000 	 9.17
95.0000 	 9.17
96.0000 	 9.17
97.0000 	 9.17
98.0000 	 9.17
99.0000 	 9.17
99.9000 	 9.17
99.9900 	 9.17
99.9990 	 9.17
99.9999 	 9.17
100.0000 	 9.17

Chapter 26 The “Tail Modulo Constructor” program transformation

(Introduced in OCaml 4.14)

Note: this feature is considered experimental, and its interface may evolve, with user feedback, in the next few releases of the language.

Consider this natural implementation of the List.map function:

let rec map f l = match l with | [] -> [] | x :: xs -> let y = f x in y :: map f xs

A well-known limitation of this implementation is that the recursive call, map f xs, is not in tail position. The runtime needs to remember to continue with y :: r after the call returns a value r, therefore this function consumes some amount of call-stack space on each recursive call. The stack usage of map f li is proportional to the length of li. This is a correctness issue for large lists on systems configured with limited stack space – the dreaded Stack_overflow exception.

# let with_stack_limit stack_limit f = let old_gc_settings = Gc.get () in Gc.set { old_gc_settings with stack_limit }; Fun.protect ~finally:(fun () -> Gc.set old_gc_settings) f ;;
val with_stack_limit : int -> (unit -> 'a) -> 'a = <fun>
# with_stack_limit 20_000 (fun () -> List.length (map Fun.id (List.init 1_000_000 Fun.id)) );;
Stack overflow during evaluation (looping recursion?).

In this implementation of map, the recursive call happens in a position that is not a tail position in the program, but within a datatype constructor application that is itself in tail position. We say that those positions, that are composed of tail positions and constructor applications, are tail modulo constructor (TMC) positions – we sometimes write tail modulo cons for brevity.

It is possible to rewrite programs such that tail modulo cons positions become tail positions; after this transformation, the implementation of map above becomes tail-recursive, in the sense that it only consumes a constant amount of stack space. The OCaml compiler implements this transformation on demand, using the [@tail_mod_cons] or [@ocaml.tail_mod_cons] attribute on the function to transform.

let[@tail_mod_cons] rec map f l = match l with | [] -> [] | x :: xs -> let y = f x in y :: map f xs
# List.length (map Fun.id (List.init 1_000_000 Fun.id));;
- : int = 1000000

This transformation only improves calls in tail-modulo-cons position, it does not improve recursive calls that do not fit in this fragment:

(* does *not* work: addition is not a data constructor *) let[@tail_mod_cons] rec length l = match l with | [] -> 0 | _ :: xs -> 1 + length xs
Warning 71 [unused-tmc-attribute]: This function is marked @tail_mod_cons but is never applied in TMC position.

It is of course possible to use the [@tail_mod_cons] transformation on functions that contain some recursive calls in tail-modulo-cons position, and some calls in other, arbitrary positions. Only the tail calls and tail-modulo-cons calls will happen in constant stack space.

General design

This feature is provided as an explicit program transformation, not an implicit optimization. It is annotation-driven: the user is expected to express their intent by adding annotations in the program using attributes, and will be asked to do so in any ambiguous situation.

We expect it to be used mostly by advanced OCaml users needing to get some guarantees on the stack-consumption behavior of their programs. Our recommendation is to use the [@tailcall] annotation on all callsites that should not consume any stack space. [@tail_mod_cons] extends the set of functions on which calls can be annotated to be tail calls, helping establish stack-consumption guarantees in more cases.

Performance

A standard approach to get a tail-recursive version of List.map is to use an accumulator to collect output elements, and reverse it at the end of the traversal.

let rec map f l = map_aux f [] l and map_aux f acc l = match l with | [] -> List.rev acc | x :: xs -> let y = f x in map_aux f (y :: acc) xs

This version is tail-recursive, but it is measurably slower than the simple, non-tail-recursive version. In contrast, the tail-mod-cons transformation provides an implementation that has comparable performance to the original version, even on small inputs.

Evaluation order

Beware that the tail-modulo-cons transformation has an effect on evaluation order: the constructor argument that is transformed into tail-position will always be evaluated last. Consider the following example:

type 'a two_headed_list = | Nil | Consnoc of 'a * 'a two_headed_list * 'a let[@tail_mod_cons] rec map f = function | Nil -> Nil | Consnoc (front, body, rear) -> Consnoc (f front, map f body, f rear)

Due to the [@tail_mod_cons] transformation, the calls to f front and f rear will be evaluated before map f body. In particular, this is likely to be different from the evaluation order of the unannotated version. (The evaluation order of constructor arguments is unspecified in OCaml, but many implementations typically use left-to-right or right-to-left.)

This effect on evaluation order is one of the reasons why the tail-modulo-cons transformation has to be explicitly requested by the user, instead of being applied as an automatic optimization.

Why tail-modulo-cons?

Other program transformations, in particular a transformation to continuation-passing style (CPS), can make all functions tail-recursive, instead of targeting only a small fragment. Some reasons to provide builtin support for the less-general tail-mod-cons are as follows:

Note: OCaml call stack size

In OCaml 4.x and earlier, bytecode programs respect the stack_limit runtime parameter configuration (as set using Gc.set in the example above), or the l setting of the OCAMLRUNPARAM variable. Native programs ignore these settings and only respect the operating system native stack limit, as set by ulimit on Unix systems. Most operating systems run with a relatively low stack size limit by default, so stack overflow on non-tail-recursive functions are a common programming bug.

Starting from OCaml 5.0, native code does not use the native system stack for OCaml function calls anymore, so it is not affected by the operating system native stack size; both native and bytecode programs respect the OCaml runtime’s own limit. The runtime limit is set to a much higher default than most operating system native stacks, with a limit of at least 512MiB, so stack overflow should be much less common in practice. There is still a stack limit by default, as it remains useful to quickly catch bugs with looping non-tail-recursive functions. Without a stack limit, one has to wait for the whole memory to be consumed by the stack for the program to crash, which can take a long time and make the system unresponsive.

This means that the tail modulo constructor transformation is less important on OCaml 5: it does improve performance noticeably in some cases, but it is not necessary for basic correctness for most use-cases.

26.1 Disambiguation

It may happen that several arguments of a constructor are recursive calls to a tail-modulo-cons function. The transformation can only turn one of these calls into a tail call. The compiler will not make an implicit choice, but ask the user to provide an explicit disambiguation.

Consider this type of syntactic expressions (assuming some pre-existing type var of expression variables):

type var (* some pre-existing type of variables *) type exp = | Var of var | Let of binding * exp and binding = var * exp

Consider a map function on variables. The direct definition has two recursive calls inside arguments of the Let constructor, so it gets rejected as ambiguous.

let[@tail_mod_cons] rec map_vars f exp = match exp with | Var v -> Var (f v) | Let ((v, def), body) -> Let ((f v, map_vars f def), map_vars f body)
Error: [@tail_mod_cons]: this constructor application may be TMC-transformed in several different ways. Please disambiguate by adding an explicit [@tailcall] attribute to the call that should be made tail-recursive, or a [@tailcall false] attribute on calls that should not be transformed. This call could be annotated. This call could be annotated.

To disambiguate, the user should add a [@tailcall] attribute to the recursive call that should be transformed to tail position:

let[@tail_mod_cons] rec map_vars f exp = match exp with | Var v -> Var (f v) | Let ((v, def), body) -> Let ((f v, map_vars f def), (map_vars[@tailcall]) f body)

Be aware that the resulting function is not tail-recursive, the recursive call on def will consume stack space. However, expression trees tend to be right-leaning (lots of Let in sequence, rather than nested inside each other), so putting the call on body in tail position is an interesting improvement over the naive definition: it gives bounded stack space consumption if we assume a bound on the nesting depth of Let constructs.

One would also get an error when using conflicting annotations, asking for two of the constructor arguments to be put in tail position:

let[@tail_mod_cons] rec map_vars f exp = match exp with | Var v -> Var (f v) | Let ((v, def), body) -> Let ((f v, (map_vars[@tailcall]) f def), (map_vars[@tailcall]) f body)
Error: [@tail_mod_cons]: this constructor application may be TMC-transformed in several different ways. Only one of the arguments may become a TMC call, but several arguments contain calls that are explicitly marked as tail-recursive. Please fix the conflict by reviewing and fixing the conflicting annotations. This call is explicitly annotated. This call is explicitly annotated.

26.2 Danger: getting out of tail-mod-cons

Due to the nature of the tail-mod-cons transformation (see Section 26.3 for a presentation of transformation):

The fact that calls in tail position in the source program may become non-tail calls if they go from a tail-mod-cons to a non-tail-mod-cons function is surprising, and the transformation will warn about them.

For example:

let[@tail_mod_cons] rec flatten = function | [] -> [] | xs :: xss -> let rec append_flatten xs xss = match xs with | [] -> flatten xss | x :: xs -> x :: append_flatten xs xss in append_flatten xs xss
Warning 71 [unused-tmc-attribute]: This function is marked @tail_mod_cons but is never applied in TMC position. Warning 72 [tmc-breaks-tailcall]: This call is in tail-modulo-cons position in a TMC function, but the function called is not itself specialized for TMC, so the call will not be transformed into a tail call. Please either mark the called function with the [@tail_mod_cons] attribute, or mark this call with the [@tailcall false] attribute to make its non-tailness explicit.

Here the append_flatten helper is not annotated with [@tail_mod_cons], so the calls append_flatten xs xss and flatten xss will not be tail calls. The correct fix here is to annotate append_flatten to be tail-mod-cons.

let[@tail_mod_cons] rec flatten = function | [] -> [] | xs :: xss -> let[@tail_mod_cons] rec append_flatten xs xss = match xs with | [] -> flatten xss | x :: xs -> x :: append_flatten xs xss in append_flatten xs xss

The same warning occurs when append_flatten is a non-tail-mod-cons function of the same recursive group; using the tail-mod-cons transformation is a property of individual functions, not whole recursive groups.

let[@tail_mod_cons] rec flatten = function | [] -> [] | xs :: xss -> append_flatten xs xss and append_flatten xs xss = match xs with | [] -> flatten xss | x :: xs -> x :: append_flatten xs xss
Warning 71 [unused-tmc-attribute]: This function is marked @tail_mod_cons but is never applied in TMC position. Warning 72 [tmc-breaks-tailcall]: This call is in tail-modulo-cons position in a TMC function, but the function called is not itself specialized for TMC, so the call will not be transformed into a tail call. Please either mark the called function with the [@tail_mod_cons] attribute, or mark this call with the [@tailcall false] attribute to make its non-tailness explicit.

Again, the fix is to specialize append_flatten as well:

let[@tail_mod_cons] rec flatten = function | [] -> [] | xs :: xss -> append_flatten xs xss and[@tail_mod_cons] append_flatten xs xss = match xs with | [] -> flatten xss | x :: xs -> x :: append_flatten xs xss

Non-recursive functions can also be annotated [@tail_mod_cons]; this is typically useful for local bindings to recursive functions.

Incorrect version:

let[@tail_mod_cons] rec map_vars f exp = let self exp = map_vars f exp in match exp with | Var v -> Var (f v) | Let ((v, def), body) -> Let ((f v, self def), (self[@tailcall]) body)
Warning 51 [wrong-tailcall-expectation]: expected tailcall Warning 51 [wrong-tailcall-expectation]: expected tailcall Warning 71 [unused-tmc-attribute]: This function is marked @tail_mod_cons but is never applied in TMC position.

Recommended fix:

let[@tail_mod_cons] rec map_vars f exp = let[@tail_mod_cons] self exp = map_vars f exp in match exp with | Var v -> Var (f v) | Let ((v, def), body) -> Let ((f v, self def), (self[@tailcall]) body)

In other cases, there is either no benefit in making the called function tail-mod-cons, or it is not possible: for example, it is a function parameter (the transformation only works with direct calls to known functions).

For example, consider a substitution function on binary trees:

type 'a tree = Leaf of 'a | Node of 'a tree * 'a tree let[@tail_mod_cons] rec bind (f : 'a -> 'a tree) (t : 'a tree) : 'a tree = match t with | Leaf v -> f v | Node (left, right) -> Node (bind f left, (bind[@tailcall]) f right)
Warning 72 [tmc-breaks-tailcall]: This call is in tail-modulo-cons position in a TMC function, but the function called is not itself specialized for TMC, so the call will not be transformed into a tail call. Please either mark the called function with the [@tail_mod_cons] attribute, or mark this call with the [@tailcall false] attribute to make its non-tailness explicit.

Here f is a function parameter, not a direct call, and the current implementation is strictly first-order, it does not support tail-mod-cons arguments. In this case, the user should indicate that they realize this call to f v is not, in fact, in tail position, by using (f[@tailcall false]) v.

type 'a tree = Leaf of 'a | Node of 'a tree * 'a tree let[@tail_mod_cons] rec bind (f : 'a -> 'a tree) (t : 'a tree) : 'a tree = match t with | Leaf v -> (f[@tailcall false]) v | Node (left, right) -> Node (bind f left, (bind[@tailcall]) f right)

26.3 Details on the transformation

To use this advanced feature, it helps to be aware that the function transformation produces a specialized function in destination-passing-style.

Recall our map example:

let rec map f l = match l with | [] -> [] | x :: xs -> let y = f x in y :: map f xs

Below is a description of the transformed program in pseudo-OCaml notation: some operations are not expressible in OCaml source code. (The transformation in fact happens on the Lambda intermediate representation of the OCaml compiler.)

let rec map f l =
  match l with
  | [] -> []
  | x :: xs ->
    let y = f x in
    let dst = y ::{mutable} Hole in
    map_dps f xs dst 1;
    dst

and map_dps f l dst idx =
  match l with
  | [] -> dst.idx <- []
  | x :: xs ->
    let y = f x in
    let dst' = y ::{mutable} Hole in
    dst.idx <- dst';
    map_dps f xs dst' 1

The source version of map gets transformed into two functions, a direct-style version that is also called map, and a destination-passing-style version (DPS) called map_dps. The destination-passing-style version does not return a result directly, instead it writes it into a memory location specified by two additional function parameters, dst (a memory block) and i (a position within the memory block).

The source call y :: map f xs gets transformed into the creation of a mutable block y ::{mutable} Hole, whose second parameter is an un-initialized hole. The block is then passed to map_dps as a destination parameter (with offset 1).

Notice that map does not call itself recursively, it calls map_dps. Then, map_dps calls itself recursively, in a tail-recursive way.

The call from map to map_dps is not a tail call (this is something that we could improve in the future); but this call happens only once when invoking map f l, with all list elements after the first one processed in constant stack by map_dps.

This explains the “getting out of tail-mod-cons” subtleties. Consider our previous example involving mutual recursion between flatten and append_flatten.

let[@tail_mod_cons] rec flatten l =
  match l with
  | [] -> []
  | xs :: xss ->
    append_flatten xs xss

The call to append_flatten, which syntactically appears in tail position, gets transformed differently depending on whether the function has a destination-passing-style version available, that is, whether it is itself annotated [@tail_mod_cons]:

(* if append_flatten_dps exists *)
and flatten_dps l dst i =
  match l with
  | [] -> dst.i <- []
  | xs :: xss ->
    append_flatten_dps xs xss dst i

(* if append_flatten_dps does not exist *)
and rec flatten_dps l dst i =
  match l with
  | [] -> dst.i <- []
  | xs :: xss ->
    dst.i <- append_flatten xs xss

If append_flatten does not have a destination-passing-style version, the call gets transformed to a non-tail call.

26.4 Current limitations

Purely syntactic criterion

Just like tail calls in general, the notion of tail-modulo-constructor position is purely syntactic; some simple refactoring will move calls out of tail-modulo-constructor position.

(* works as expected *) let[@tail_mod_cons] rec map f li = match li with | [] -> [] | x :: xs -> let y = f x in y :: (* this call is in TMC position *) map f xs
(* not optimizable anymore *) let[@tail_mod_cons] rec map f li = match li with | [] -> [] | x :: xs -> let y = f x in let ys = (* this call is not in TMC position anymore *) map f xs in y :: ys
Warning 71 [unused-tmc-attribute]: This function is marked @tail_mod_cons but is never applied in TMC position.
Local, first-order transformation

When a function gets transformed with tail-mod-cons, two definitions are generated, one providing a direct-style interface and one providing the destination-passing-style version. However, not all calls to this function in tail-modulo-cons position will use the destination-passing-style version and become tail calls:

Consider the call Option.map foo x for example: even if foo is called in tail-mod-cons position within the definition of Option.map, that call will never become a tail call. (This would be the case even if the call to Option.map was inside the Option module.)

In general this limitation is not a problem for recursive functions: the first call from an outside module or a higher-order function will consume stack space, but further recursive calls in tail-mod-cons position will get optimized. For example, if List.map is defined as a tail-mod-cons function, calls from outside the List module will not become tail calls when in tail positions, but the recursive calls within the definition of List.map are in tail-modulo-cons positions and do become tail calls: processing the first element of the list will consume stack space, but all further elements are handled in constant space.

These limitations may be an issue in more complex situations where mutual recursion happens between functions, with some functions not annotated tail-mod-cons, or defined across different modules, or called indirectly, for example through function parameters.

Non-exact calls to tupled functions

OCaml performs an implicit optimization for “tupled” functions, which take a single parameter that is a tuple: let f (x, y, z) = .... Direct calls to these functions with a tuple literal argument (like f (a, b, c)) will call the “tupled” function by passing the parameters directly, instead of building a tuple of them. Other calls, either indirect calls or calls passing a more complex tuple value (like let t = (a, b, c) in f t) are compiled as “inexact” calls that go through a wrapper.

The [@tail_mod_cons] transformation supports tupled functions, but will only optimize “exact” calls in tail position; direct calls to something other than a tuple literal will not become tail calls. The user can manually unpack a tuple to force a call to be “exact”: let (x, y, z) = t in f (x, y, z). If there is any doubt as to whether a call can be tail-mod-cons-optimized or not, one can use the [@tailcall] attribute on the called function, which will warn if the transformation is not possible.

let rec map (f, l) = match l with | [] -> [] | x :: xs -> let y = f x in let args = (f, xs) in (* this inexact call cannot be tail-optimized, so a warning will be raised *) y :: (map[@tailcall]) args
Warning 51 [wrong-tailcall-expectation]: expected tailcall

Part IV
The OCaml library

Chapter 27 The core library

This chapter describes the OCaml core library, which is composed of declarations for built-in types and exceptions, plus the module Stdlib that provides basic operations on these built-in types. The Stdlib module is special in two ways:

27.1 Built-in types and predefined exceptions

The following built-in types and predefined exceptions are always defined in the compilation environment, but are not part of any module. As a consequence, they can only be referred by their short names.

Built-in types

 type int

The type of integer numbers.
 type char

The type of characters.
 type bytes

The type of (writable) byte sequences.
 type string

The type of (read-only) character strings.
 type float

The type of floating-point numbers.
 type bool = false | true

The type of booleans (truth values).
 type unit = ()

The type of the unit value.
 type exn

The type of exception values.
 type 'a array

The type of arrays whose elements have type 'a.
 type 'a list = [] | :: of 'a * 'a list

The type of lists whose elements have type 'a.
type 'a option = None | Some of 'a

The type of optional values of type 'a.
type int32

The type of signed 32-bit integers. Literals for 32-bit integers are suffixed by l. See the Int32 module.
type int64

The type of signed 64-bit integers. Literals for 64-bit integers are suffixed by L. See the Int64 module.
type nativeint

The type of signed, platform-native integers (32 bits on 32-bit processors, 64 bits on 64-bit processors). Literals for native integers are suffixed by n. See the Nativeint module.
type ('a, 'b, 'c, 'd, 'e, 'f) format6

The type of format strings. 'a is the type of the parameters of the format, 'f is the result type for the printf-style functions, 'b is the type of the first argument given to %a and %t printing functions (see module Printf), 'c is the result type of these functions, and also the type of the argument transmitted to the first argument of kprintf-style functions, 'd is the result type for the scanf-style functions (see module Scanf), and 'e is the type of the receiver function for the scanf-style functions.
type 'a lazy_t

This type is used to implement the Lazy module. It should not be used directly.

Predefined exceptions

exception Match_failure of (string * int * int)

Exception raised when none of the cases of a pattern-matching apply. The arguments are the location of the match keyword in the source code (file name, line number, column number).
exception Assert_failure of (string * int * int)

Exception raised when an assertion fails. The arguments are the location of the assert keyword in the source code (file name, line number, column number).
exception Invalid_argument of string

Exception raised by library functions to signal that the given arguments do not make sense. The string gives some information to the programmer. As a general rule, this exception should not be caught, it denotes a programming error and the code should be modified not to trigger it.
exception Failure of string

Exception raised by library functions to signal that they are undefined on the given arguments. The string is meant to give some information to the programmer; you must not pattern match on the string literal because it may change in future versions (use Failure _ instead).
exception Not_found

Exception raised by search functions when the desired object could not be found.
exception Out_of_memory

Exception raised by the garbage collector when there is insufficient memory to complete the computation. (Not reliable for allocations on the minor heap.)
exception Stack_overflow

Exception raised by the bytecode interpreter when the evaluation stack reaches its maximal size. This often indicates infinite or excessively deep recursion in the user’s program. Before 4.10, it was not fully implemented by the native-code compiler.
exception Sys_error of string

Exception raised by the input/output functions to report an operating system error. The string is meant to give some information to the programmer; you must not pattern match on the string literal because it may change in future versions (use Sys_error _ instead).
exception End_of_file

Exception raised by input functions to signal that the end of file has been reached.
exception Division_by_zero

Exception raised by integer division and remainder operations when their second argument is zero.
exception Sys_blocked_io

A special case of Sys_error raised when no I/O is possible on a non-blocking I/O channel.
exception Undefined_recursive_module of (string * int * int)

Exception raised when an ill-founded recursive module definition is evaluated. (See section 12.2.) The arguments are the location of the definition in the source code (file name, line number, column number).

27.2 Module Stdlib: the initially opened module

Chapter 28 The standard library

This chapter describes the functions provided by the OCaml standard library. The modules from the standard library are automatically linked with the user’s object code files by the ocamlc command. Hence, these modules can be used in standalone programs without having to add any .cmo file on the command line for the linking phase. Similarly, in interactive use, these globals can be used in toplevel phrases without having to load any .cmo file in memory.

Unlike the core Stdlib module, submodules are not automatically “opened” when compilation starts, or when the toplevel system is launched. Hence it is necessary to use qualified identifiers to refer to the functions provided by these modules, or to add open directives.

Chapter 29 The compiler front-end

This chapter describes the OCaml front-end, which declares the abstract syntax tree used by the compiler, provides a way to parse, print and pretty-print OCaml code, and ultimately allows one to write abstract syntax tree preprocessors invoked via the -ppx flag (see chapters 13 and 16).

It is important to note that the exported front-end interface follows the evolution of the OCaml language and implementation, and thus does not provide any backwards compatibility guarantees.

The front-end is a part of compiler-libs library. Programs that use the compiler-libs library should be built as follows:

        ocamlfind ocamlc other options -package compiler-libs.common other files
        ocamlfind ocamlopt other options -package compiler-libs.common other files

Use of the ocamlfind utility is recommended. However, if this is not possible, an alternative method may be used:

        ocamlc other options -I +compiler-libs ocamlcommon.cma other files
        ocamlopt other options -I +compiler-libs ocamlcommon.cmxa other files

For interactive use of the compiler-libs library, start ocaml and type
#load "compiler-libs/ocamlcommon.cma";;.

Chapter 30 The unix library: Unix system calls

The unix library makes many Unix system calls and system-related library functions available to OCaml programs. This chapter describes briefly the functions provided. Refer to sections 2 and 3 of the Unix manual for more details on the behavior of these functions.

Not all functions are provided by all Unix variants. If some functions are not available, they will raise Invalid_arg when called.

Programs that use the unix library must be linked as follows:

        ocamlc other options -I +unix unix.cma other files
        ocamlopt other options  -I +unix unix.cmxa other files

For interactive use of the unix library, do:

        ocamlmktop -o mytop -I +unix unix.cma
        ./mytop

or (if dynamic linking of C libraries is supported on your platform), start ocaml and type

# #directory "+unix";;
# #load "unix.cma";;
Windows:  The Cygwin port of OCaml fully implements all functions from the Unix module. The native Win32 ports implement a subset of them. Below is a list of the functions that are not implemented, or only partially implemented, by the Win32 ports. Functions not mentioned are fully implemented and behave as described previously in this chapter.
FunctionsComment
forknot implemented, use create_process or threads
waitnot implemented, use waitpid
waitpidcan only wait for a given PID, not any child process
getppidnot implemented (meaningless under Windows)
nicenot implemented
truncate, ftruncateimplemented (since 4.10.0)
linkimplemented (since 3.02)
fchmodnot implemented
chown, fchownnot implemented (make no sense on a DOS file system)
umasknot implemented
accessexecute permission X_OK cannot be tested, it just tests for read permission instead
chrootnot implemented
mkfifonot implemented
symlink, readlinkimplemented (since 4.03.0)
killpartially implemented (since 4.00.0): only the sigkill signal is implemented
sigprocmask, sigpending, sigsuspendnot implemented (no inter-process signals on Windows
pausenot implemented (no inter-process signals in Windows)
alarmnot implemented
timespartially implemented, will not report timings for child processes
getitimer, setitimernot implemented
getuid, geteuid, getgid, getegidalways return 1
setuid, setgid, setgroups, initgroupsnot implemented
getgroupsalways returns [|1|] (since 2.00)
getpwnam, getpwuidalways raise Not_found
getgrnam, getgrgidalways raise Not_found
type socket_domainPF_INET is fully supported; PF_INET6 is fully supported (since 4.01.0); PF_UNIX is supported since 4.14.0, but only works on Windows 10 1803 and later.
establish_servernot implemented; use threads
terminal functions (tc*)not implemented
setsidnot implemented

Chapter 31 The str library: regular expressions and string processing

The str library provides high-level string processing functions, some based on regular expressions. It is intended to support the kind of file processing that is usually performed with scripting languages such as awk, perl or sed.

Programs that use the str library must be linked as follows:

        ocamlc other options -I +str str.cma other files
        ocamlopt other options  -I +str str.cmxa other files

For interactive use of the str library, do:

        ocamlmktop -o mytop str.cma
        ./mytop

or (if dynamic linking of C libraries is supported on your platform), start ocaml and type

# #directory "+str";;
# #load "str.cma";;

Chapter 32 The runtime_events library

The runtime_events library provides an API for consuming runtime tracing and metrics information from the runtime. See chapter 25 for more information.

Programs that use runtime_events must be linked as follows:

        ocamlc -I +runtime_events other options unix.cma runtime_events.cma other files
        ocamlopt -I +runtime_events other options unix.cmxa runtime_events.cmxa other files

Compilation units that use the runtime_events library must also be compiled with the -I +runtime_events option (see chapter 13).

Chapter 33 The threads library

The threads library allows concurrent programming in OCaml. It provides multiple threads of control (also called lightweight processes) that execute concurrently in the same memory space. Threads communicate by in-place modification of shared data structures, or by sending and receiving data on communication channels.

The threads library is implemented on top of the threading facilities provided by the operating system: POSIX 1003.1c threads for Linux, MacOS, and other Unix-like systems; Win32 threads for Windows. Only one thread at a time is allowed to run OCaml code on a particular domain 9.5.1. Hence, opportunities for parallelism are limited to the parts of the program that run system or C library code. However, threads provide concurrency and can be used to structure programs as several communicating processes. Threads also efficiently support concurrent, overlapping I/O operations.

Programs that use threads must be linked as follows:

        ocamlc -I +unix -I +threads other options unix.cma threads.cma other files
        ocamlopt -I +unix -I +threads other options unix.cmxa threads.cmxa other files

Compilation units that use the threads library must also be compiled with the -I +threads option (see chapter 13).

Chapter 34 The dynlink library: dynamic loading and linking of object files

The dynlink library supports type-safe dynamic loading and linking of bytecode object files (.cmo and .cma files) in a running bytecode program, or of native plugins (usually .cmxs files) in a running native program. Type safety is ensured by limiting the set of modules from the running program that the loaded object file can access, and checking that the running program and the loaded object file have been compiled against the same interfaces for these modules. In native code, there are also some compatibility checks on the implementations (to avoid errors with cross-module optimizations); it might be useful to hide .cmx files when building native plugins so that they remain independent of the implementation of modules in the main program.

Programs that use the dynlink library simply need to include the dynlink library directory with -I +dynlink and link dynlink.cma or dynlink.cmxa with their object files and other libraries.

Note: in order to insure that the dynamically-loaded modules have access to all the libraries that are visible to the main program (and not just to the parts of those libraries that are actually used in the main program), programs using the dynlink library should be linked with -linkall.

Chapter 35 Recently removed or moved libraries (Graphics, Bigarray, Num, LablTk)

This chapter describes three libraries which were formerly part of the OCaml distribution (Graphics, Num, and LablTk), and a library which has now become part of OCaml’s standard library, and is documented there (Bigarray).

35.1 The Graphics Library

Since OCaml 4.09, the graphics library is distributed as an external package. Its new home is:

https://github.com/ocaml/graphics

If you are using the opam package manager, you should install the corresponding graphics package:

        opam install graphics

Before OCaml 4.09, this package simply ensures that the graphics library was installed by the compiler, and starting from OCaml 4.09 this package effectively provides the graphics library.

35.2 The Bigarray Library

As of OCaml 4.07, the bigarray library has been integrated into OCaml’s standard library.

The bigarray functionality may now be found in the standard library Bigarray module, except for the map_file function which is now part of the Unix library. The documentation has been integrated into the documentation for the standard library.

The legacy bigarray library bundled with the compiler is a compatibility library with exactly the same interface as before, i.e. with map_file included.

We strongly recommend that you port your code to use the standard library version instead, as the changes required are minimal.

If you choose to use the compatibility library, you must link your programs as follows:

        ocamlc other options bigarray.cma other files
        ocamlopt other options bigarray.cmxa other files

For interactive use of the bigarray compatibility library, do:

        ocamlmktop -o mytop bigarray.cma
        ./mytop

or (if dynamic linking of C libraries is supported on your platform), start ocaml and type #load "bigarray.cma";;.

35.3 The Num Library

The num library implements integer arithmetic and rational arithmetic in arbitrary precision. It was split off the core OCaml distribution starting with the 4.06.0 release, and can now be found at https://github.com/ocaml/num.

New applications that need arbitrary-precision arithmetic should use the Zarith library (https://github.com/ocaml/Zarith) instead of the Num library, and older applications that already use Num are encouraged to switch to Zarith. Zarith delivers much better performance than Num and has a nicer API.

35.4 The Labltk Library and OCamlBrowser

Since OCaml version 4.02, the OCamlBrowser tool and the Labltk library are distributed separately from the OCaml compiler. The project is now hosted at https://github.com/garrigue/labltk.

Part V
Indexes

Index of keywords


This document was translated from LATEX by HEVEA.