English version
Accueil     À propos     Téléchargement     Ressources     Contactez-nous    

Ce site est rarement mis à jour. Pour les informations les plus récentes, rendez-vous sur le nouveau site OCaml à l'adresse ocaml.org.

Browse thread
[ANN] Vprint: a runtime value printer module
[ Home ] [ Index: by date | by threads ]
[ Search: ]

[ Message by date: previous | next ] [ Message in thread: previous | next ] [ Thread: previous | next ]
Date: 2007-08-24 (01:18)
From: Zheng Li <li@p...>
Subject: [ANN] Vprint: a runtime value printer module


In the spirit of "release early, release often", we therefore announce the
first release of Vprint, a runtime value printer module for OCaml. Be warned,
the module is extremely experimental at the moment, and the implementation is
very hasty (I got the main idea yesterday when finalising my GSoc project and
now it's here). So be ready to encounter stupid errors and even core dump, but
do remember to send me a message about that so I can improve. Here is the
introduction from README:


Vprint is a value printer module for OCaml. It prints any value at running time
with a simple generic printer. It can be used for the following purposes: help
debugging, inspect data representation, test type casting, runtime dispatch
based on type representations and avoid writing pretty printers.

  - Help debugging. 

    For rapid development reason, we don't want to write a printer/formatter
    function for each data type. Moreover, even if we've got the printers for
    each type, specifying printers to inspect the running time values, which
    are usually the arbitrary combinations (tuple, list, variants, objects
    etc.) of them, is still tedious. 

  - Inspect data representation

    The module can shows you all kinds of data (of different types) that having
    the same memory layout as a input data. 

  - Test type casting

    By printing a value with the format of another type, it can help to test
    whether a data can be safely cast to another type without actually doing
    it. But be warned, the real casting can still fail even if the printing

  - Runtime dispatch based on type representations

    In the same spirit as above.

  - Avoid writing pretty printers (TODO)

    Without considering debugging, sometimes you still need to output the
    values of some data type. If you don't care much about the format and just
    want the result shew, then you probably don't have to write it by your own
    now; if the OCaml toplevel printer is something you'd like, we're on the
    way there, if given more time.


There are just two main functions "print" and "print_all" (and their
output-to-string version "sprint" and "sprint_all")

    val print: ?fmt:fmt -> 'a -> unit
    val print_all: ?fmt:fmt -> 'a -> unit

The "print" function can prints any data with a generic printer. However, it's
well-known fact that OCaml runtime doesn't carry type information, so the
output maybe too generic to read sometimes, e.g. the representation of list is
something like a linked pairs. 

The implementation tries to inference the actual data type, however due to the
theoretical limitation, there do exist many cases where the output is not
satisfactory such as the list/pair example. And we don't believe it's
appropriate to just choose a probable result based on probabilities, because
misleading is even worse. (1, (2, (3, (4, (5, (6, 0)))))) is mostly likely to
be a list, but not necessarily.

In such cases, you can give more hints to the system by using the optional ~fmt
parameters. The format of fmt is very close to type representation and supports
combination, so it's quite easy to write (check example section). Besides it
accepts the default arguments "__" (means "anything") and try its best to
inference more information based on the less information being given, so you
don't have to write every details. 

In such as sense, fmt can be used as a simpler formatter if you want to pretty
print some values as output but is lazy to write complex formatter, just gives
more specific fmt information if you want better effect. For now, we use rather
simple string output function, but we'll definitely move to Format and build
better output a la OCaml's toplevel output.

You can also use "print" to do preliminary tests of type casting. By printing
value x of type t with a fmt associated with type t', you get more confidence
to cast x to type t', instead of making real casting and get core dump, here
you only get a Fmt_Error exception. I believe you've been warned it's not
definitive, but the point is that the function can help you to rule out a lot
possibilities beforehand.

In the same spirit, you can use the function to do runtime dispatch based on
type representations (not types! it's like representation level duck
typing). E.g, something like (not tested yet)

    open Vprint
    let o = input_value ch in
    try ignore (sprint ~fmt:_s o); do_sth_to_o_as_string
    with Fmt_Error _ ->
      try ignore (sprint ~fmt:_f o); do_sth_to_o_as_float
      with Fmt_Error _ ->
        try ignore (sprint ~fmt:_l_ o); do_sth_to_o_as_list
        with Fmt_Error _ -> ......

Finally, the "print_all" function can help you to inspect the representation of
data, especially it lists all the data (of any types) that having the same
memory representation as a value. For example, when you print a value with the
default formatter: "print v", the output is so generic that won't satisfy
you. Then you may wonder why the output is so generic? You can get the answer
with "print_all v", then you'll see all kinds of values which having the same
memory representation as your input "v", that's why the system won't be able to
decide which one of them is the "v" hence print less meaningful result.


It's a simple module, just compile the source to cmo/cmx to use.  The source
itself makes use of camlp4 (just for stream function), so compile it with
option "-pp camlp4o". The result modules doesn't depend on camlp4, so you don't
have to specify the camlp4 option when you make use of vprint.

We also provide Makefile script with  the following command: make all, make
install, make doc, make clean, make uninstall. 

Only tested under the following environment: Linux, OCaml 3.10.0, GNU Make 3.81.


See file ./LICENCE


# open Vprint

(* Some values are easy to identify *)
# print ("asdf", [||], 3.14, [|1.01; 2.689|], 32l, 64L, 111n);;
<"asdf", [||], 3.14, [|1.01; 2.689|], 32l, 64L, 111n>

(* Some are difficult *)
# print [1; 2; 3; 4];;
<1', <2', <3', <4', 0'>>>>

(* By giving it some type information, it's much better. 
   "_l __" means list of 'any, or use _l_ instead *)
# print ~fmt:(_l __) [1; 2; 3; 4];;
[1'; 2'; 3'; 4']

(* Why the number has a ' after it? *)
# print 0;;

(* Because there are several datum having this same representations we don't
   know who is this guy currently *)
# print_all 0;;
[]       (* it could be empty list *)
()       (* or the unit *)
'^@'     (* or the char '^@" *)
Con0     (* or the 1st no-param constructor of some variant type *)
`Var0    (* or a no-param polymorphic variants with interval id = 1 *)
0        (* or, at last, the int value 0! *)

(* print_all can also take ~fmt arguments to restrict the choices, now the ' is
   gone since 0 is the only choice as a int "_i" *)
# print_all ~fmt:_i 0;;

(* Like the "'", "< >" is sequence whose type is yet to decide *)
# print (3, 4);;
<3', 4'>
- : unit = ()
# print_all (3, 4);;
[|'^C'; '^D'|]        (* array of 2 char, first is '^C", then '^D' *)
{'^C'; '^D'}          (* record of 2 fields, both char *)
('^C', '^D')          (* tuple of 2 char *)
Con0# ('^C', '^D')    (* it's a variant type, here the value is its 1st
                         with-param constructors with 2 char as params *)
('^C', Con4)          (* tuple of char and variant whose value here is its 5th
                         no-param constructor *)
`Var3 4               (* a polymorphic variant type, its value here is a
                         constructor with id=3 and taking int 4 as param *)
[|3; 4|]              (* int array of two int *)
(3, 4)                (* int tuple, our current input! *)
......                (* 50+ possibilities in total *)

(* Really ? test *)
# type t = A of char * char | B | C of t list;;
# print (A ('\003', '\004'), [|3; 4|]);;
<<3', 4'>, <3', 4'>>

(* Make use of ~fmt, check manual for details, 
   be aware about the 'any: "_" "__" *)

(* Print polymorphic variants *)
# print ~fmt:(_l _'v_) [`How "vvv"; `Are 3.333; `You];;
[`How "vvv"; `Are 3.333; `You]

(* Print functions, code * environment *)
# let k = 5;;
# let rec f x = g x + k and g x = f x - k;;
# print f; print g;;
<fun#67404426 5'>
<fun#67404440 5'>

(* Print object, id * code * environment *)
# print (object val x = 10 val y = 15 method get_xy = x + y end);;
<obj#0/67328382 (10', 15')>

(* Print forced lazy value *)
# let l = lazy (8.28, "vvv") in let _ = Lazy.force l in print l;;
lazy <8.28, "vvv">

(* fmt = 'any * int, *~ is the AND combinator, for collection *)
# print ~fmt:(__ *~ _i) (3.14, 3);;
<3.14, 3>

(* fmt = 'v list and 'v = C' of char * char | C'' of 'any list 
   The // is OR combinator, for variants and polymorphic variants.
   You can rewrite the variants part as _v (_c *~ _c) // _v _l_ *)
# print ~fmt: (_l (_v (_c *~ _c // _l_)))  [A ('x', 'y'); C []];;
[Con0# ('x', 'y'); Con1# []]

(* fmt = int * 'any list * 'b array and 'b = 'any variant *)
# print ~fmt:(_i *~ _l_ *~ _a _v_) (3, [3.14; 9.9], [|B; B|]);;
<3, [3.14; 9.9], [|Con0; Con0|]>

(* fmt = int array list list, the use of combinator @. (read as "OF" as in
   "list of list of int array) can reduce "..))))..", otherwise we should
   write it as (_l (_l (_a _i)))  *)
# print ~fmt:(_l @._l @._a _i) [[[|1; 2|]; [|3; 4|]]];;
[[[|1; 2|]; [|3; 4|]]]

(* Anyway, it's really hard to describe all the details here, you'll have to
   read the manual and play with it by yourself to understand. *)

Zheng Li