Re: [PATCH v2] Fortran: Narrow return types [PR78798]

[Mikael Morin <morin-mikael@orange.fr>]

Le 14/05/2023 à 17:24, Bernhard Reutner-Fischer a écrit :
> On Sun, 14 May 2023 15:04:15 +0200
> Thomas Koenig <tkoenig@netcologne.de> wrote:
> 
>> On 14.05.23 14:27, Mikael Morin wrote:
>>>
>>> (...)
>>>> @@ -2098,7 +2098,7 @@ ref_same_as_full_array (gfc_ref *full_ref,
>>>> gfc_ref *ref)
>>>>            there is some kind of overlap.
>>>>        0 : array references are identical or not overlapping.  */
>>>> -int
>>>> +bool
>>>>    gfc_dep_resolver (gfc_ref *lref, gfc_ref *rref, gfc_reverse *reverse,
>>>>              bool identical)
>>>>    {
>>>
>>> The function comment states that the function may return 2, which
>>> doesn't seem to be the case any more.
>>
>> Hm, this makes me a litte suspicious.  Was functionality for reversing
>> loops lost,  maybe unintentionally?  I assume that, at some time, we did
>> use the '2' as return value (or did we?)
> 
> There was 7c428aa29d75ef163c334cf3974f87b3630d8b8b (a revert because it
> miscompiled spec2k) which might have associated the comment of the
> former static gfc_dependency dep_ref (gfc_ref *lref, gfc_ref *rref,
> gfc_reverse *reverse) to the current gfc_dep_resolver.
> 
> The commit which introduced the return value 2 documentation was
> 3d03ead0b8273efde57f6194617b35111a84b05d
> "re PR fortran/24524 (Fortran dependency checking should reverse loops)"
> 
> but TBH i don't see how it returned 2 in that revision?
> Looks like when writing that patch it deemed useful to return 2 for
> this specific situation but in the end it was dropped but the comment
> survived.
>
Yes, I came to the same conclusion that we never returned 2 here.

The information that reversal is needed is already provided on a per 
dimension basis by the gfc_reverse pointer passed as argument, so 
providing the information in the return value would be redundant anyway.


> I will update the comment to document the true / false return values.
> 
> And Mikael, do you want me to cleanup 1/0 to true/false assignments for
> the boolean variables, or can we do that in a separate patch (or not at
> all right now)?
> 
I don't mind too much either way.
As long as the variables are not assigned integer values outside of the 
[0,1] range, and we consistently use true/false or 0/1, not a mix of 
them, it's fine with me.