Thread: BUG #18870: weird behavior with regexp_replace

The following bug has been logged on the website:

Bug reference:      18870
Logged by:          reinko
Email address:      devops@key2asset.com
PostgreSQL version: 17.4
Operating system:   Ubuntu 11.4.0-1ubuntu1~22.04
Description:

select regexp_replace(LOWER('Örebro'), '\W', '_', 'g')  in postgres 15 the
result is örebro which is correct since ö should fit in the \w for a
regex.
select regexp_replace(LOWER('Örebro'), '\W', '_', 'g') --> since postgres 17
the result is _rebro which is incorrect since \w should also contain
characters like ö, ä, ë.

the to lower is not really relevant to this issue the same happens when it's
just a direct string aswell.
this issue happens with alot of special a-z characters é, è have the same
issue for example.

Kind regards,
Reinko Brink

PG Bug reporting form <noreply@postgresql.org> writes:
> select regexp_replace(LOWER('Örebro'), '\W', '_', 'g')  in postgres 15 the
> result is örebro which is correct since ö should fit in the \w for a
> regex.
> select regexp_replace(LOWER('Örebro'), '\W', '_', 'g') --> since postgres 17
> the result is _rebro which is incorrect since \w should also contain
> characters like ö, ä, ë.

This most likely indicates that you've got a different database
collation selected in the v17 installation.  Postgres defers to
the LC_CTYPE setting (or, in some configurations, the ICU collation)
to decide what is a letter.  See

https://www.postgresql.org/docs/current/charset.html

psql's "\l" command will give a quick overview of what collations
you have selected.

            regards, tom lane