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>]

Hi Thomas,

>> I tested it on a fairly large code base and found no further false
>> positives. Also it still regtests cleanly. Ok for trunk?
>
>
> while I still disagree with this on principle, I will not stand
> in the way.

IIUC you disagree in the sense that you would like gfortran to have
more such optimizations (with more corresponding warnings). Is that
correct?

I hope you can at least agree that the warnings I'm adding in the
patch are a) useful and b) sufficient for the current state of
optimizations?

Modifying gfortran's runtime behavior is really a separate question
that we can continue to discuss later, probably with some amount of
controversy. But before we even go there, I would really like to have
warnings for the optimizations we have now ...


> However, one point: I think that the warning should be under a separate
> warning, which should then be enabled by -Wextra.
> -Waggressive-function-elimination, could be reused for this,
> or something else

I don't actually see such a flag in the manual.

I do see "-faggressive-function-elimination"
(https://gcc.gnu.org/onlinedocs/gfortran/Code-Gen-Options.html), but
that is about changing runtime behavior, not about throwing warnings.

Do you propose to couple the warning to
-faggressive-function-elimination (and also the short-circuiting
behavior itself)?
Or do you propose to add a flag named -Waggressive-function-elimination?

Cheers,
Janus