Andrey Borodin wrote:
> Overall patch looks solid and thoughtful work and adds important
> functionality.
I tried the patch, with some minor changes to build with HEAD.
I was surprised by the interface, that is, the fact that a user is
not allowed to freely choose the ICU collation of a database, in
constrast with CREATE COLLATION.
AFAIU, when the "default collation provider" is ICU, CREATE DATABASE
still expects a libc locale in the lc_collate/lc_ctype arguments.
The code will automatically find an ICU equivalent by matching the
language, and it seems that the country is ignored?
So if we wanted a database with an ICU collation like, say,
"es@collation=traditional" or "es-u-co-trad" as expressed with a
BCP-47 tag, or anything that is not defined by only a language,
would it be possible? I have the impression it wouldn't.
This is not something that could be easily improved after the fact
because getting the ICU collation through a libc collation is a
user-interface choice.
I think users would rather be able to create a database with
something like:
CREATE DATABASE foo
COLLPROVIDER='icu'
LOCALE='icu_locale' |
LC_COLLATE='icu_locale' | LC_CTYPE='icu_locale'
...
which would be in line with CREATE COLLATION.
Best regards,
--
Daniel Vérité
PostgreSQL-powered mailer: http://www.manitou-mail.org
Twitter: @DanielVerite