Re: xlc 12.1 miscompiles 32-bit ginCompareItemPointers()

On Sun, Jul 19, 2015 at 04:22:47PM -0400, Tom Lane wrote:
> Andres Freund <andres(at)anarazel(dot)de> writes:
> > On July 19, 2015 9:50:33 PM GMT+02:00, Noah Misch <noah(at)leadboat(dot)com> wrote:
> >> I propose to expand the gin_private.h "#ifdef PG_USE_INLINE" test to
> >> exclude xlc 32-bit configurations. The last 32-bit AIX kernel exited
> >> support on 2012-04-30.
>
> > I vote to simply error out in that case then. Trying to fix individual compiler bugs in an niche OS sounds like a bad idea.

Fair principle. PostgreSQL 9.4.4 does work in this configuration due to the
accident of -qlonglong causing a warning that in turn disables PG_USE_INLINE.
I do not want 9.4.5 to break this configuration, and I also don't want to
restore the works-accidentally state. Adding a few more conditions to one #if
is cheap and has neither problem.

> I think I'm with Andres --- are there really enough users of this
> configuration to justify working around such a bug?

We never have the benefit of an answer to that question.

> More: if the compiler does have a bug like that, how much confidence can
> we have, really, that there are no other miscompiled places and won't be
> any in the future? If we are going to support this configuration, I think
> what the patch ought to do is disable PG_USE_INLINE globally when this
> compiler is detected. That would revert the behavior to what it was
> before 43d89a2.

That's a reasonable alternative.