Tim Tassonis wrote:
Hi
I just recently compiled and installed 8.3.1 on a System that has UTF-8
as the default characterset in the environment. Copied the binaries, run
initdb without parameters, the usual stuff.
As you probably are all aware of, this results now in a cluster that
will only allow you to create UTF-8 databases. I have read some posts
regarding this topic where it is explained that allowing LATIN1 on a
cluster initialized with UTF-8 will give you problems anyway etc and you
have to use locale=C to be able to create databases with different
charactersets.
You can only have one locale per installation (initdb'd cluster).
You can create a database with any encoding that is compatible with that
locale (LATIN1, LATIN9, UTF-8, etc).
Now, the locale controls (amongst other things) all your sorting. If you
choose a locale of "C" you get a simple binary sorting. By default PG
tries to match whatever sorting you have set up on your operating-system.
See the difference below:
richardh@server3:db$ LANG=C sort /tmp/words.txt
apple
berry
Apple
apple
apples
the apple
richardh@server3:db$ LANG=en_GB.UTF-8 sort /tmp/words.txt
apple
apple
Apple
apples
berry
the apple
I think someone is looking at per-database locales for 8.4 - the issue
is more tricky than you might think because you need to worry about
system catalogue sort-order.
--
Richard Huxton
Archonet Ltd
