Re: [Caml-list] Problem with the compiler 3.12.1

[Claire Dross <claire.dross@lri.fr>]

On 25/04/2012 10:04, Claire Dross wrote:
> On 16/04/2012 17:52, Claire Dross wrote:
>> Hello,
>>
>> I am new to caml-list, so sorry in advance if it is not the right 
>> place to make this post. In a quite important piece of ocaml code, I 
>> have a function:
>>
>> let matching_term terms env uf id (pats, subst) (seen, forms) =
>>     match pats, subst with
>>       | _, [] | [], _ -> seen, forms
>>       | t :: pats, _ -> fprintf fmt "   %a@." T.print t;
>>         let acc = matchpats env terms uf (subst, seen, id+1) t in
>>         let acc, seen, _ = List.fold_left (matchpats env env.fils uf)
>>           acc pats in
>>         seen, List.rev_append acc forms
>>
>> Removing the print in the second case of the match changes the result 
>> of the compiled file (compiler 3.12.1 byte code and for linux)) on 
>> one of my tests. T.print does not have side effects, it recursively 
>> prints a hashconsed term structure.
>>
> Using the compiler 3.11.2 instead of the 3.12.1 removes the problem. 
> My function returns the correct result with and without the print.
>
> Thanks again to everyone that took time to help me,
>
> Claire Dross
>
The problem is caused by the garbage collector (launched because Format 
creates a lot of things) that introduces non-determinism in the order of 
elements in a weak hash table.