Re: Crash report for some ICU-52 (debian8) COLLATE and work_mem values

On Mon, Aug 14, 2017 at 11:32 AM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> Hmm ... this all seems like good stuff, but it also seems like new
> development, and it's way too late in the v10 cycle for any significant
> amount of that.

Fair enough.

> I think at this point I'd vote to drop auto-installation of keyword
> variants, and I would also vote against Peter's auto_comment proposal.
> We shouldn't be adding features that we're going to supersede in v11.

+1

> Maybe, if people are okay with a catversion bump this late, we could
> push the ICU descriptions into pg_collation proper, but I think it
> would be fine to leave pg_import_system_collations's behavior in that
> regard alone for v10, too.

I really think we should add a pg_collation column to store ICU's
description of the collation, because that's something that we'll have
to live with forever.

I don't think we have pg_import_system_collations's behavior all
worked out just yet. I'm not sure that we'd do the right thing at
initdb even if pg_import_system_collations did not add the keyword
variants. It would still add all distinct collations, instead of all
CLDR locales (some of which will have equivalent collation behavior).
I suspect that this explains the stability problems you saw [1] that
were not explainable by available variants differing.

I do think that we need to have a better answer to this "locales vs.
collations" question for v10.

[1] postgr.es/m/CAH2-Wzm3EOEqYB48m7aT7sjPXw=kUHU=FdzzdU90dfiatt62CQ@mail.gmail.com

--
Peter Geoghegan