On Mon, Sep 25, 2017 at 12:14 PM, Peter Geoghegan <pg@bowt.ie> wrote:
> But then our users categorically have to know about both formats,
> without any practical benefit to make up for it. You will also get
> people that don't realize that only one format is supported on some
> versions if go this way.
Oh, and if you go this way there is also a much bigger disadvantage:
you also break pg_restore when you cross the ICU 54 version boundary
in *either* direction (pg_collation.collname will be different, so
actually equivalent collations will not be recognized as such for
dump/restore purposes). This seems very likely, even, because ICU 54
isn't that old (we support versions as old as ICU 4.2), and because
you only have to have used ICU initdb collation for this to happen (no
CREATE COLLATION is necessary). Sure, you *may* be able to first do a
manual CREATE COLLATION when that happens, and that will be picked up
during the restore. But that's very user hostile.
That must have been the real reason why you canonicalized
pg_collation.collname (I doubt it had anything to do with how keyword
variants used to be created during initdb, as you suggested). As Tom
pointed out recently, we've actually always canonicalized collation
name for libc.
--
Peter Geoghegan
--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers