Re: [Caml-list] Strange performances

[Benjamin Canou <benjamin.canou@gmail.com>]

  Hi,

Indeed, using runtime exceptions catching as a programming style may
lead to strange behaviours, but I think using the any pattern in the try
with construct was the real mistake.

So basically, to those who did not understand the problem :
This code works perfectly :

let list_of_string s =
  let rec list_of_string s i =
    try let e = s.[i] in e :: list_of_string s (succ i)
    with Invalid_argument "index out of bounds" -> []
  in list_of_string s 0

And if I had used a correct matching of exceptions in my original code :

let list_of_string s =
  let rec list_of_string s i =
    try s.[i] :: list_of_string s (succ i)
    with Invalid_argument "index out of bounds" -> []
  in list_of_string s 0

I would have noticed that the bug came from my carelessness about the
evaluation order (btw, thank you Jacques) :

Fatal error: exception Stack_overflow

In fact, the function calls itself recursively to construct the right
hand side of the list constructor until the stack is full.

With the any (_) pattern, the function returns [] for each call from the
one corresponding to the end of the string to the one causing the
overflow. So the result is correct, but it takes a time proportional to
the maximum size of the stack...
With a pattern matching the out of bounds exception, the stack overflow
is not caught and the programs exits abnormally right after the
overflow.

Jacques, if I remember well, the ocaml runtime is not able to detect
stack overflows in native code on all platforms, that's why you get a
segfault instead of a Stack overflow exception.

  Benjamin Canou.

Le vendredi 18 janvier 2008 à 18:12 +0900, Jacques Garrigue a écrit :
> From: "Till Varoquaux" <till.varoquaux@gmail.com>
> > On Jan 18, 2008 2:15 AM, Jacques Garrigue <garrigue@math.nagoya-u.ac.jp> wrote:
> > ...
> > > By the way, on my machine your version doesn't even work in native
> > > code, I only get segfaults. This is allowed behaviour for
> > > out-of-bounds access.
> > 
> > Could you please clarify? This seems a little scary to me, I thought
> > segfaults where acceptable only when you used unsafe features (or ran
> > out of stack).
> 
> This is why I sent an erratum. The cause for the segfault was not the
> array access, but the stack overflow, which occured due to ocaml's
> peculiar evaluation order.
> Still, I maintain that intentionally raising and catching out-of-bound
> accesses is not good programming style...
> 
> Jacques Garrigue
> 
> _______________________________________________
> Caml-list mailing list. Subscription management:
> http://yquem.inria.fr/cgi-bin/mailman/listinfo/caml-list
> Archives: http://caml.inria.fr
> Beginner's list: http://groups.yahoo.com/group/ocaml_beginners
> Bug reports: http://caml.inria.fr/bin/caml-bugs