Horton

Re: pg_trgm comparison bug on cross-architecture replication due to different char implementation

Masahiko Sawada <sawada.mshk@gmail.com>

View thread View on postgr.es
From: Masahiko Sawada <sawada.mshk@gmail.com>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: Noah Misch <noah@leadboat.com>, Joe Conway <mail@joeconway.com>, Peter Eisentraut <peter@eisentraut.org>, Alexander Korotkov <aekorotkov@gmail.com>, "Guo, Adam" <adamguo@amazon.com>, "pgsql-hackers@lists.postgresql.org" <pgsql-hackers@lists.postgresql.org>, Nathan Bossart <nathandbossart@gmail.com>, Jim Mlodgenski <jimmy76@gmail.com>
Date: 2024-10-03T13:55:47Z
Lists: pgsql-hackers

Attachments

On Wed, Oct 2, 2024 at 10:02 AM Masahiko Sawada <sawada.mshk@gmail.com> wrote:
>
> On Tue, Oct 1, 2024 at 8:57 PM Tom Lane <tgl@sss.pgh.pa.us> wrote:
> >
> > Noah Misch <noah@leadboat.com> writes:
> > > On Tue, Oct 01, 2024 at 11:55:48AM -0700, Masahiko Sawada wrote:
> > >> Considering that the population of database cluster signedness will
> > >> converge to signedness=true in the future, we can consider using
> > >> -fsigned-char to prevent similar problems for the future. We need to
> > >> think about possible side-effects as well, though.
> >
> > > It's good to think about -fsigned-char.  While I find it tempting, several
> > > things would need to hold for us to benefit from it:
> >
> > > - Every supported compiler has to offer it or an equivalent.
> > > - The non-compiler parts of every supported C implementation need to
> > >   cooperate.  For example, CHAR_MIN must change in response to the flag.  See
> > >   the first comment in cash_in().
> > > - Libraries we depend on can't do anything incompatible with it.
> >
> > > Given that, I would lean toward not using -fsigned-char.  It's unlikely all
> > > three things will hold.  Even if they do, the benefit is not large.
> >
> > I am very, very strongly against deciding that Postgres will only
> > support one setting of char signedness.  It's a step on the way to
> > hardware monoculture, and we know where that eventually leads.
> > (In other words, I categorically reject Sawada-san's assertion
> > that signed chars will become universal.  I'd reject the opposite
> > assertion as well.)
>
> Thank you for pointing this out. I agree with both of you.
>

I've attached PoC patches for the idea Noah proposed. Newly created
clusters unconditionally have default_char_signedness=true, and the
only source of signedness=false is pg_upgrade. To update the
signedness in the controlfile, pg_resetwal now has a new option
--char-signedness, which is used by pg_upgrade internally. Feedback is
very welcome.

Regards,

--
Masahiko Sawada
Amazon Web Services: https://aws.amazon.com

Commits

  1. pg_upgrade: Check for the expected error message in TAP tests.

  2. Fix a typo in 005_char_signedness.pl test.

  3. Add test 005_char_signedness.pl to meson.build.

  4. Fix an issue with index scan using pg_trgm due to char signedness on different architectures.

  5. pg_upgrade: Add --set-char-signedness to set the default char signedness of new cluster.

  6. pg_upgrade: Preserve default char signedness value from old cluster.

  7. pg_resetwal: Add --char-signedness option to change the default char signedness.

  8. Add default_char_signedness field to ControlFileData.

  9. Remove unneeded nbtree array preprocessing assert.