Horton

Re: Collation versioning

Thomas Munro <thomas.munro@enterprisedb.com>

View thread View on postgr.es
From: Thomas Munro <thomas.munro@enterprisedb.com>
To: Peter Eisentraut <peter.eisentraut@2ndquadrant.com>
Cc: Christoph Berg <myon@debian.org>, Pg Hackers <pgsql-hackers@postgresql.org>
Date: 2018-09-07T00:36:15Z
Lists: pgsql-hackers 
On Thu, Sep 6, 2018 at 12:01 PM Peter Eisentraut
<peter.eisentraut@2ndquadrant.com> wrote:
> On 05/09/2018 23:18, Thomas Munro wrote:
> > On Wed, Sep 5, 2018 at 12:10 PM Christoph Berg <myon@debian.org> wrote:
> >>> So, it's not ideal but perhaps worth considering on the grounds that
> >>> it's better than nothing?
> >>
> >> Ack.
> >
> > Ok, here's a little patch like that.
> >
> > postgres=# select collname, collcollate, collversion from pg_collation
> > where collname = 'en_NZ';
> >  collname | collcollate | collversion
> > ----------+-------------+-------------
> >  en_NZ    | en_NZ.utf8  | 2.24
> > (1 row)
>
> But wouldn't that also have the effect that glibc updates that don't
> change anything about the locales would trigger the version
> incompatibility warning?

Right.  And likewise, a glibc update that does change some locales but
not the locales that you are actually using will trigger false alarm
warnings. The same goes for the ICU provider, which appears to return
the same collversion for every collation, even though presumably some
don't change from one ICU version to the next.

I wonder if someone here knows how many "locales" packages have been
released over the lifetime of (say) the current Debian stable distro,
whether any LC_COLLATE files changed over those releases, and whether
libc6 had the same MAJOR.MINOR for the whole lifetime.  That is, even
though they might have been through 2.19-17+blah, 2.19-18+blah, ...
did they all report "2.19" and were the collations actually stable?
If that's the case, I think it'd be quite good: we'd only raise the
alarm after a big dist-upgrade Debian 8->9, or when doing streaming
replication from a Debian 8 box to a Debian 9 box.

> Also, note that this mechanism only applies to collation objects, not to
> database-global locales.  So most users wouldn't be helped by this approach.

Yeah, right, that would have to work for this to be useful.  I will
look into that.

--
Thomas Munro
http://www.enterprisedb.com

Commits

  1. Doc: Document known problem with Windows collation versions.

  2. Add collation versions for FreeBSD.

  3. Tolerate version lookup failure for old style Windows locale names.

  4. Track collation versions for indexes.

  5. Add pg_depend.refobjversion.

  6. Remove pg_collation.collversion.

  7. Fix the MSVC build for versions 2015 and later.

  8. Add collation versions for Windows.

  9. Implement type regcollation

  10. Use libc version as a collation version on glibc systems.

  11. Make type "name" collation-aware.