Re: [Patch, Fortran] PR 85599: warn about short-circuiting of logical expressions for non-pure functions

[Janus Weil <janus at gcc dot gnu dot org>]

2018-07-12 16:35 GMT+02:00 Dominique d'Humières <dominiq@lps.ens.fr>:
>
>>>> after the dust of the heated discussion around this PR has settled a
>>>> bit, here is another attempt to implement at least some basic warnings
>>>> about compiler-dependent behavior concerning the short-circuiting of
>>>> logical expressions. …
>>>
>>> IMO your patch is missing the only point I agree with you on this issue, i.e.,
>>
>> So you don't agree that it's a good idea to warn about non-portable code?
>
> Wel, as you might guess I don’t like warnings in general: too much false positives
> and missed targets (see recent bootstrap breakage).

So then: Do you have a better solution for the problem at hand?

The cleaner approach would certainly be to avoid short-circuiting of
impure functions altogether. If we can all agree that this is a good
idea, I'll be happy to submit a patch in that direction. But recent
discussions have shown that there is a broad opposition of hardliners
who favor optimization to portability. I don't really have the nerves
and the time to continue this endless fight, so for now I would be
satisfied if we at least had the warnings.


>>> What to do with
>>>
>>> if(x>0 .and. sort(x)<10.0*log(x)) …
>>>
>>> which is not portable to compilers computing the two expressions?
>>
>> That's one of the typical cases that the patch should be able to
>> handle. If 'sort' is impure, this should trigger a warning with the
>> patch. If it is pure, it doesn't matter whether the rhs of the .and.
>> operator is executed (since there are no side effects). Or am I
>> missing something?
>
> I meant ‘sqrt’ (changed behind my back by the stupid spell-checker).

sqrt and log are both elemental, thus pure, so there will be no warning.

Cheers,
Janus