Re: Suggestion for a moratorium on short-circuit IF

Re: Suggestion for a moratorium on short-circuit IF

[Dominique d'Humières]

> after some thoughts and off-line discussion with another contributor,
> I would like to propose a moratorium on this topic.
> …

I fully agree.

Dominique

Suggestion for a moratorium on short-circuit IF

[Thomas Koenig]

Hi,

after some thoughts and off-line discussion with another contributor,
I would like to propose a moratorium on this topic.

It seems that we are unable to agree to a course of action, or even
the direction that the course should take us.  Also, I think that
a continuing disagreement could do more harm to gfortran development
than this topic actually merits.

So, my proposal would be to freeze the current status, which is better
than gcc 8 because we have added a warning that was not there before,
and because we now have the documentation that users can turn to.

I would also like to thank Janus for the work he has put into this,
and the other participants of the discussion.

Maybe we can discuss this over a beer at the GNU cauldron (hint, hint).
I'll buy a round.

Regards

	Thomas

Re: Suggestion for a moratorium on short-circuit IF

[Janus Weil]

Hi Thomas,

sorry, but I don't really understand why you're making such a fuzz
about this issue.

Sure, the discussion of the whole topic has been a bit lengthy and
quite controversial (and part of the blame certainly goes to the
Fortran standard for not being more stringent in this area). But in
the end I would say that the process was quite useful. It has helped
to illuminate the issue from all angles and brought up all relevant
boundary conditions.

From my point of view the debate has crystallized into a very clear
end result: Short-circuiting is not mandatory in Fortran, but it is a
useful and important optimization. My proposal (and in fact two other
people have proposed this before me) was to treat it like every other
optimization and enable it with -O flags (providing the best possible
performance), but disable it with -O0 (in order to allow for debugging
and error diagnostics).

It seems that for some reason you have a diffuse aversion to this
approach, but you haven't really explained why.

I would strongly prefer to have a factual debate, instead of somebody
killing off the topic as non-debatable. Please explain what the
problem is. I think I'm a reasonable person and open to objective
arguments. However, I don't understand at all why this topic is such a
holy cow and afflicted with so much dogma.

So, a simple question to you personally: Why, in your opinion, is it
so utterly important to do short-circuiting at -O0? I really see no
good reason why we should.

Cheers,
Janus

PS: Regarding the beers-at-the-cauldron thing ... that really is an
excellent idea and I was actually trying to make it happen, but
unfortunately there are severe personal-life constraints that make it
hard if not impossible for me to come to Manchester in September
(which is not quite around the corner from where I live).



2018-07-25 23:23 GMT+02:00 Thomas Koenig <tkoenig@netcologne.de>:
> Hi,
>
> after some thoughts and off-line discussion with another contributor,
> I would like to propose a moratorium on this topic.
>
> It seems that we are unable to agree to a course of action, or even
> the direction that the course should take us.  Also, I think that
> a continuing disagreement could do more harm to gfortran development
> than this topic actually merits.
>
> So, my proposal would be to freeze the current status, which is better
> than gcc 8 because we have added a warning that was not there before,
> and because we now have the documentation that users can turn to.
>
> I would also like to thank Janus for the work he has put into this,
> and the other participants of the discussion.
>
> Maybe we can discuss this over a beer at the GNU cauldron (hint, hint).
> I'll buy a round.
>
> Regards
>
>         Thomas