Martijn van Oosterhout napsal(a):
> On Mon, Sep 22, 2008 at 06:11:04PM +0300, Heikki Linnakangas wrote:
>>> This patch should allow to use both system catalog and ICU.
>> Not without another patch that actually introduces ICU support. What
>> that would look like, how that would be stored in the catalogs, and
>> whether we want that is whole another topic. Without that, the STRCOLFN
>> part of the original patch is pointless, and I would've ripped that out
>> anyway even if we decided to add the pg_collation catalog in this release.
>
> Eh? How you store collations names is easy. Every collation has a
> textual name which is what we store in the catalog. I'm not sure why
> you'd think it'd be any more complicated than that. And it has
> precisely nothing to do with ICU and everything to do with being able
> to support multiple source of collation information. We already have
> two: builtin (strcmp/memcmp) and C library (strcoll). Letting peope add
> more is the goal.
pg_collation catalog is also important for pg_dump, because system
collation names are not compatible over OS and pg_dump output should be
portable. pg_collation adds abstract layer which solve this problem.
> I'm sure once the catalog exists the existing ICU-for-Postgres patch
> will be adjusted to use it.
Yes, I agree with Martijn.
Zdenek
