Обсуждение: Sorting Problem in UNICODE/german

Hi there!

I have a Problem with a DB that was created in UNICODE

* createdb -E UNICODE

and actually shows that it _is_ in UNICODE.
i was able to input data and can read it and everything is fine.
but when i want to "ORDER BY ..." it does not sort the german Umlauts at the
correct postition.

should be:
m n o ö p

and is:
ö a b c d

I have tried starting postgres with LC_ALL=de_AT.utf8@euro
locale but that did not help.

what xould i do?

regs,
klaus

Klaus Ita <postgres@stro.at> writes:
> I have tried starting postgres with LC_ALL=de_AT.utf8@euro
> locale but that did not help.

You need to run initdb under that setting.  See "Localization" in
the documentation.

            regards, tom lane

On Thu, Sep 01, 2005 at 09:30:15AM -0400, Tom Lane wrote:
> Klaus Ita <postgres@stro.at> writes:
> > I have tried starting postgres with LC_ALL=de_AT.utf8@euro
> > locale but that did not help.
>

i did read the docs and am still not quite happy with my sorting results.
ok initdb has been rerun

made sure, i had the locale:

locale -a

created new db-cluster with
LC_ALL=de_AT.utf8@euro initdb --locale=de_AT.utf8@euro -E UNICODE -D /dev/shm/pgutf8

and then still the sorting was not right when i restored another UNICODE db.

another "funny" thing is:

ita@aipc54:~/.mutt$ LC_ALL=de_AT.utf8@euro sort /tmp/testfile
Abend
Oma
Ãterreich
Ãerflieger
Unter
Zetrix

this is also wrong (There should be 'Unter' and then 'U:berflieger' [Überflieger]). so is this a libc bug?


thank you for your help so far! I more than appreciate it. Support for this
DB is sooo much better than for oracle!

klaus

Klaus Ita schrob:

> On Thu, Sep 01, 2005 at 09:30:15AM -0400, Tom Lane wrote:
>> Klaus Ita <postgres@stro.at> writes:
>> > I have tried starting postgres with LC_ALL=de_AT.utf8@euro
>> > locale but that did not help.
>>
>
> i did read the docs and am still not quite happy with my sorting results.
> ok initdb has been rerun
>
> made sure, i had the locale:
>
> locale -a
>
> created new db-cluster with
> LC_ALL=de_AT.utf8@euro initdb --locale=de_AT.utf8@euro -E UNICODE -D /dev/shm/pgutf8
>
> and then still the sorting was not right when i restored another
> UNICODE db.

Well, I used the very same command with 8.0.3 to create a database,
and the sort order was correct:

--8<---------------cut here---------------start------------->8---
scratch=# select w from w order by w;
      w
-------------
 Abend
 Oma
 Österreich
 Überflieger
 Unter
 Zetrix
(6 rows)
--8<---------------cut here---------------end--------------->8---

So I guess there was some misconfiguration of your current
client_encoding during import, or maybe the dump of your unicode db
got unexpectedly converted by improper settings during dumping.

> another "funny" thing is:
>
> ita@aipc54:~/.mutt$ LC_ALL=de_AT.utf8@euro sort /tmp/testfile
> Abend
> Oma
> Ãterreich
> Ãerflieger
> Unter
> Zetrix
>
> this is also wrong (There should be 'Unter' and then 'U:berflieger'
> [Überflieger]). so is this a libc bug?

The sort order is correct, so libc did succeed in its part. Maybe your
terminal is having issues with utf-8? If you're using xterm: Did you
run it with -u8 or some utf-8-enabling X-resource? To verify that the
terminal is working properly, typing

    echo ö > /tmp/foo
    file /tmp/foo

on a shell should tell you that you have a utf-8 text file.

HTH
Andreas
--

Sorry, I just reread your mail: Your MUA is declaring it with

> Content-Type: text/plain; charset=unknown-8bit

This makes it even harder to discuss problems with Umlauts :-).

Andreas Seltenreich schrob:

> Klaus Ita schrob:
>
>> another "funny" thing is:
>>
>> ita@aipc54:~/.mutt$ LC_ALL=de_AT.utf8@euro sort /tmp/testfile
>> Abend
>> Oma
>> Ãterreich
>> Ãerflieger
>> Unter
>> Zetrix
>>
>> this is also wrong (There should be 'Unter' and then 'U:berflieger'
>> [Überflieger]). so is this a libc bug?

I think I got your point now. Libc appears to be using iso-14651
sorting for all "de" locales. I'm afraid you will have compile a
customized locale to depart from that.

regards,
Andreas
--

Andreas Seltenreich <andreas+pg@gate450.dyndns.org> writes:
>> Klaus Ita schrob:
>>> this is also wrong (There should be 'Unter' and then 'U:berflieger'
>>> [Überflieger]). so is this a libc bug?

> I think I got your point now. Libc appears to be using iso-14651
> sorting for all "de" locales. I'm afraid you will have compile a
> customized locale to depart from that.

I wouldn't call it a libc bug, but a bug in the locale definition.
In any case it doesn't appear to be Postgres' problem --- if we sort
the same way "sort" does under the same locale setting, then we are
doing what we expect.

I think at this point Klaus needs to find some people who know about
hacking locale definitions.  I sure don't know enough about them to
help further.  Is there a libc mailing list anywhere?

One thing I do know --- if you install a new version of the locale
Postgres is using, you'd better re-initdb, or at least REINDEX all
your indexes on textual columns.  Changing sort order is equivalent
to making such indexes corrupt.

            regards, tom lane