Thanks. I hoped for something possible inside a pl/pgsql proc. I'm
trying to integrate pg_trgm with Tsearch2. I'm still on my UTF-8
database. Yes I know, there is _NO_ UTF-8 support of any kind in
Tsearch2 yet, but I got it working to a degree that is OK for my
application (Created my own stemmer variant, ispell dict, affix file
etc). The last missing bit is to get a source for pg_trgm. I cannot use
the the stat() function, because it breaks as soon it sees an UTF-8 char.
I suppose noncompatible with UTF wordparser can produce illegal lexemes (with
part of multibyte char) and stores it in tsvector. Tsvector hasn't any control
of breakness lexemes (with a help pg_verifymbstr() call), but stat() makes text
field and then postgres check it and found incomplete mbchars. Which way I see
(except waiting UTF support in tsearch2 which we develop now):
1 modify stat() function to check text field and if it fails then remove lexeme
from output
2 Take from CVS HEAD wordpaser (ts_locale.[ch], wparser_def.c,
wordparser/parser.[ch]). to_tsvector will works fine, to_tsquery will works
correct only with quoted string (for examle, 'foo' & 'bar', bad: foo & bar).
But casting 'asasas'::tsvector and dump/reload will not work correct.
--
Teodor Sigaev E-mail: teodor@xxxxxxxxx
WWW: http://www.sigaev.ru/
