Version française
Home     About     Download     Resources     Contact us    
Browse thread
Parameter evaluation order
[ Home ] [ Index: by date | by threads ]
[ Search: ]

[ Message by date: previous | next ] [ Message in thread: previous | next ] [ Thread: previous | next ]
Date: -- (:)
From: Márk_S._Zoltán <zoltan.s.mark@d...>
Subject: Parameter evaluation order
I was recently thinking about parameter evaluation order and how come it 
is unspecified in OCaml. Maybe this was already discussed here, but I do 
not recall it and cannot find it, either. Do not get me wrong, I do not 
have an unsurmontable problem with unspecified parameter evaluation 
ordering, and in a world full of languages with similar disclaimers I 
could not even have a problem. Also, side effects in parameters is just 
ugly. So the current shape of the language is fine with me.

But I do think that currying + strictness + side-effects => 
left-to-right evaluation order.  That is because a multi-parameter 
function application should be functionally equivalent to a closure 
taking the first (== leftmost) parameter and returning a closure which 
takes the second parameter etc. while parameters should be evaluated 
before passing, as per strictness. 

Consider the following code:

==== ordertest.ml =====
let f a b c d = ()

let () =
  let ff1 = f (print_string "1") in
  let ff2 = ff1 (print_string "2") in
  let ff3 = ff2 (print_string "3") in
  ff3 (print_string "4");
  print_newline ()

let () =
  let f2 = f (print_string "1") (print_string "2")
  in f2 (print_string "3") (print_string "4") ;
  print_newline ()

let () =
  f
    (print_string "1")
    (print_string "2")
    (print_string "3")
    (print_string "4");
  print_newline ()
==============

The three let ()'s should be functionally equivalent, and yet they are not:

---------------------
$ ocaml ordertest.ml
1234
2143
4321
---------------------

ocamlopt creates a different outcome, but that is well known: the last 
line reads 1234.

Of course, if this kind of logic was applied, labels and FFI's would be 
difficult to handle. Probably even enrichment of the type system with 
'has side effect' and forbidding the use of expressions with such types 
in parameters would be simpler than that, not that I advocate either 
solution. I do not say we should have left-to-right evaluation order, 
only that it would be more logical.  Am I wrong?

    M-