strange problem

=?iso-8859-1?Q?=22B=F6hm=2C_Sebastian_=28Vendor=29=22?= <sebastian.boehm@friendscout24.de> writes:

ici=3D# select id,pseudonym from user_all where pseudonym =3D
'autologin_funkey';
id | pseudonym
----+-----------
(0 rows)

ici=3D# select id,pseudonym from user_all where pseudonym ~
'^autologin_funkey$';
id | pseudonym
------+------------------
6881 | autologin_funkey
(1 row)

That's pretty bizarre. Can you give a procedure for reproducing this
from a standing start?

regards, tom lane

Hi,

I did reindex, vacuum , dump/import, ...
I also tried 7.3.1

-->> It happens only with btree index on that column, without index it
works, with hash index also no problem.

This happend to 0.3% of all rows in this table, after dump/import exactly
the same rows were affected.

Can somebody tell me how to import a 7.3(.1) dump into 7.2.3 ?
It gives errors on import, so I cannot try it on 7.2.3.

Hope that helps.

Sebastian Böhm
Friendscout24

-----Ursprüngliche Nachricht-----
Von: Tom Lane [mailto:tgl@sss.pgh.pa.us]
Gesendet: Mittwoch, 18. Dezember 2002 17:42
An: "Böhm, Sebastian (Vendor)"
Cc: 'pgsql-bugs@postgresql.org'
Betreff: Re: [BUGS] strange problem

=?iso-8859-1?Q?=22B=F6hm=2C_Sebastian_=28Vendor=29=22?=
<sebastian.boehm@friendscout24.de> writes:

ici=3D# select id,pseudonym from user_all where pseudonym =3D
'autologin_funkey';
id | pseudonym
----+-----------
(0 rows)

ici=3D# select id,pseudonym from user_all where pseudonym ~
'^autologin_funkey$';
id | pseudonym
------+------------------
6881 | autologin_funkey
(1 row)

That's pretty bizarre. Can you give a procedure for reproducing this
from a standing start?

regards, tom lane

=?iso-8859-1?Q?=22B=F6hm=2C_Sebastian_=28Vendor=29=22?= <sebastian.boehm@friendscout24.de> writes:

This happend to 0.3% of all rows in this table, after dump/import exactly
the same rows were affected.

So you can reproduce the problem after dumping/reloading? If you could
send me the dump file (off-list!) I'd be willing to investigate here ...

regards, tom lane

=?iso-8859-1?Q?=22B=F6hm=2C_Sebastian_=28Vendor=29=22?= <sebastian.boehm@friendscout24.de> writes:

I installed an newer rpm of the glibc (2.2.2-8.1 / 2.2.2-4 before) und
reinitialized the database with locale=3DC.

Just after installing the rpm, the problem was still there (reindex I =
did),
but after dump/initdb --locale=3DC/import the problem was gone.
Dont know wheater it is the reinit with new glibc or the reinit with
locale=3DC, dont have the time to figure it out.

If I am correct in guessing that strcoll is to blame, then setting the
locale to C would have made the problem go away. I am dubious that
glibc 2.2.2-8 fixes the strcoll bug --- the notes I have say that that
fix appeared in glibc 2.2.3. However, if you're happy running in C
locale then it doesn't really matter whether strcoll works or not ...

regards, tom lane

Hi,

I installed an newer rpm of the glibc (2.2.2-8.1 / 2.2.2-4 before) und
reinitialized the database with locale=C.

Just after installing the rpm, the problem was still there (reindex I did),
but after dump/initdb --locale=C/import the problem was gone.
Dont know wheater it is the reinit with new glibc or the reinit with
locale=C, dont have the time to figure it out.
(have to sleep now)

Thank you for your help !

Sebastian Böhm

-----Ursprüngliche Nachricht-----
Von: Tom Lane [mailto:tgl@sss.pgh.pa.us]
Gesendet: Donnerstag, 19. Dezember 2002 00:18
An: "Böhm, Sebastian (Vendor)"
Betreff: Re: AW: [BUGS] strange problem

Thanks for sending the test data. Unsurprisingly, I don't see any
failure here:

regression=# select id,pseudonym from test4 where pseudonym =
'autologin_funkey';
id | pseudonym
------+------------------
6881 | autologin_funkey
(1 row)

but I was already suspecting a platform-dependent or locale-dependent
problem.

Please update to glibc 2.2.3 or later, then rebuild (or at least
reindex) your tables, and let me know if you can still see a problem.

regards, tom lane