Re: [HACKERS] ORDER BY different locales

On Thu, Feb 26, 2004 at 09:16:03AM -0500, Tom Lane wrote:
> Karel Zak <zakkr(at)zf(dot)jcu(dot)cz> writes:
> > I think possible solution is special function used ORDER BY clause
> > which knows to switch by safe way to wanted locales, convert string by
> > strxfrm() and switch back to backend locales.
>
> This function breaks the whole backend if an elog() failure occurs while

Fixed by sigsetjmp(Warn_restart..). I hope it's more safe now.

> it's got the wrong locale set. I believe it would also be remarkably
> slow --- doesn't setlocale() involve reading a new locale definition

You're right, it's slow. But sometimes is more important that it works
and not all queries work with thousands records like my test below.

> I think the ultimate solution to our multi-locale problems will have to
> involve abandoning the C library's support functions and writing locale
> support that allows multiple locale-defining structures referenced by

Agree. But as you said it's huge task and I think if it won't implement
in 7.5 we can add nls_string() to the contrib tree. BTW, nls_string()
is "product" of Czech database list where Oracle users have still
problems with PostgreSQL ;-)

Latest version:
ftp://ftp2.zf.jcu.cz/users/zakkr/pg/postgresql-nls-string-0.52.tar.gz

Note, I add "CC:" to pgsql-general, maybe it's interesting for some
normal users too.

Tests:

# SELECT count(*) FROM nlstest;
count
--------
100000

# SELECT data FROM nlstest ORDER BY upper(data) DESC LIMIT 1;
Time: 1213.87 ms

# SELECT data FROM nlstest ORDER BY nls_string(data, 'en_US') LIMIT 1;
Time: 4269.00 ms

Karel

--
Karel Zak <zakkr(at)zf(dot)jcu(dot)cz>
http://home.zf.jcu.cz/~zakkr/