Re: Full text search prefix matching

Heikki Rauhala <heikki.rauhala@reaktor.fi> writes:
> Should text search prefixes work predicatably as documented in [1] even if the lexemes are shorter than the query?
Howcan I get it to work? 

I believe what you're seeing can be explained by these observations:

regression=# select to_tsvector('finnish', 'sofia');
 to_tsvector
-------------
 'sof':1
(1 row)

regression=# select to_tsquery('finnish','sofia:*');
 to_tsquery
------------
 'sof':*
(1 row)

regression=# select to_tsquery('finnish','sofi:*');
 to_tsquery
------------
 'sofi':*
(1 row)

regression=# select to_tsquery('finnish','sof:*');
 to_tsquery
------------
 'sof':*
(1 row)

What this shows is that the finnish configuration includes a word-stemming
rule that strips off "ia".  It won't strip off just "i" though, so "sofi"
doesn't get reduced to the same root and therefore doesn't match "sofia".
The "*" addition does nothing for you here since it allows matching in
the other direction (query shorter than target).

I know nothing of Finnish so I can't say just how correct these particular
stemming rules are for that language; perhaps they need adjustment.  But
it seems to me that if you want blind non-language-aware prefix matching,
you probably don't want the full-text-search machinery at all.  Full text
search is meant to deal with words, both in the documents and the queries.
You might take a look at pg_trgm as an alternative.

            regards, tom lane