Re: Failures with gcd functions with GCC snapshots GCC and -O3 (?) - Mailing list pgsql-hackers

From Thomas Munro
Subject Re: Failures with gcd functions with GCC snapshots GCC and -O3 (?)
Date
Msg-id CA+hUKG+4vJ8-U6_GHyw1ppn7pFvrjFMNGbP-ohP6WgUAkV4S8A@mail.gmail.com
Whole thread Raw
In response to Re: Failures with gcd functions with GCC snapshots GCC and -O3 (?)  (Tom Lane <tgl@sss.pgh.pa.us>)
Responses Re: Failures with gcd functions with GCC snapshots GCC and -O3 (?)  (Fabien COELHO <coelho@cri.ensmp.fr>)
List pgsql-hackers
On Sat, Jun 19, 2021 at 9:46 AM Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Fabien COELHO <coelho@cri.ensmp.fr> writes:
> >> It'd sure be nice if seawasp stopped spamming the buildfarm failure log,
> >> too.
>
> > There was a silent API breakage (same API, different behavior, how nice…)
> > in llvm main that Andres figured out, which will have to be fixed at some
> > point, so this is reminder that it is still a todo…
>
> If it were *our* todo, that would be one thing; but it isn't.

Over on the other thread[1] we learned that this is an API change
affecting reference counting semantics[2], so unless there is some
discussion somewhere about reverting the LLVM change that I'm unaware
of, I'm guessing we're going to need to change our code sooner or
later.  I have a bleeding edge LLVM on my dev machine, and I'm willing
to try to reproduce the crash and write the trivial patch (that is,
figure out the right preprocessor incantation to detect the version or
feature, and bump the reference count as appropriate), if Andres
and/or Fabien aren't already on the case.

[1] https://www.postgresql.org/message-id/flat/CA%2BhUKGLEy8mgtN7BNp0ooFAjUedDTJj5dME7NxLU-m91b85siA%40mail.gmail.com
[2] https://github.com/llvm/llvm-project/commit/c8fc5e3ba942057d6c4cdcd1faeae69a28e7b671



pgsql-hackers by date:

Previous
From: Fabien COELHO
Date:
Subject: Re: Failures with gcd functions with GCC snapshots GCC and -O3 (?)
Next
From: Thomas Munro
Date:
Subject: Re: pgbench logging broken by time logic changes