Stanislav Sukholet <ctac@osib.so-cdu.ru> writes:
>> Can't reproduce this here. Â What locale settings are you using in the
>> database? Â (Particularly lc_ctype and lc_messages)
> mydb=> SHOW client_encoding ;
> client_encoding
> -----------------
> KOI8
> (1 запиÑÑ)
> mydb=> show LC_CTYPE;
> lc_ctype
> -------------
> ru_RU.koi8r
> (1 запиÑÑ)
> mydb=> show LC_MESSAGES;
> lc_messages
> -------------
> ru_RU.koi8r
> (1 запиÑÑ)
> mydb=> CREATE TABLE a (b INTEGER PRIMARY KEY);
> ERROR: ignoring unconvertible UTF-8 character 0xd3cf
OK, with that I can reproduce it in 7.4, but more recent releases
produce a bunch of "WARNING: ignoring unconvertible UTF-8 character"
notices and then complete the operation successfully.
This is basically the same problem discussed in this thread:
http://archives.postgresql.org/pgsql-patches/2005-08/msg00037.php
namely that gettext() converts the translated error message to the
encoding implied by LC_CTYPE ... but the error reporting machinery
expects the string to be in the encoding specified for the database.
I have applied a minor tweak to the 7.4 branch to make it behave more
like the later releases, ie you get a WARNING not an ERROR. However
this is certainly not really a solution --- the only reason the behavior
isn't worse is that the ru_RU message catalog doesn't try to translate
"ignoring unconvertible UTF-8 character" and so you don't get into the
recursive failure discussed in the above thread.
The bottom line is that this is one of several reasons why it's a bad
idea to use a database encoding that's incompatible with the underlying
locale settings. I doubt that we'll really be able to fix that until
we replace all our dependence on the C library's locale facilities
... which is something that will probably happen someday, but don't
hold your breath waiting :-(
In short, if you want to use UTF8 database encoding, specify a
UTF8-based locale setting when you initdb. Don't try to change
the database encoding via -E.
regards, tom lane