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

On Wed, Aug 9, 2017 at 11:46 AM, Peter Geoghegan <pg@bowt.ie> wrote:
> The base locale names, which include regional variants like Austrian
> German, are as stable as possible. pg_import_system_collations()
> should just add those.

I just noticed that ICU collations within pg_collation do not include
entries for each of the many regional variants of English, such as
English from Ireland, Britain, the Cook Islands, etc. OTOH, glibc does
have all these variants listed within pg_collation, at least on my
system. This is because we don't actually add one pg_collation entry
per ICU locale. Rather, we add one entry per distinct ICU collation,
via ucol_countAvailable() + ucol_getAvailable(). Maybe Austrian German
actually is sorted in a slightly different way to German German, and
so gets its own pg_collation entry?

I guess this is fine, because English is in practice sorted in exactly
the same way throughout all English locales (except for a few true
variants, like "English (United States, Computer)"). These unlisted
locales do have different currency symbols and so on, but that's not
something we'll use ICU for, so it's fine. This difference will need
to be documented, though. It would help if CREATE COLLATION left new
ICU collations with the same useful "Description" as initdb created
collations will have; maybe that should be added.

-- 
Peter Geoghegan