On Wed, 12 May 2004, M. Bastin wrote:
> > > And how can I do an initdb so that sorting on Unicode will work for
> >> French, Greek, Japanase, etc. users of a single database?
> >
> >AFAIK, you can't really at this time. With an appropriately crafted
> >locale, you could probably get reasonably close, but I've never actually
> >tried to work with creating one so I don't know what's involved. And, if
> >two languages had different rules for two characters you'd not be
> >supporting both.
>
> Thanks Stephan! I've found my list of locales. It's a pity only one
> language can be used at a time but as you say there are conflicting
> rules anyway.
>
> The docs say there is a speed penalty on using locales. Does anyone
> have any idea on how severe this is? I'm wondering wether I should
I'm not an expert really, but since you're already doing unicode I think
it's not going to be major with the one caveat that if you're doing LIKE
queries, you should look at the Operator Classes section of the
documentation about the *_pattern_ops operator classes.
> use the translate() function after all because of this. It would
> solve multilingual issues to a certain level and there wouldn't be a
> speed penalty since the indexes would be build on the translate()
> function too.
The translate version would presumably work for cases where you want
multiple characters to sort to the same position, but if you want say an
accented A to follow a regular A I think it might be difficult to
formulate.