Comment 2 for bug 1187433

Revision history for this message
Dan Scott (denials) wrote :

Confirmed on our 2.4 test server, too; we're still running 2.3 in production and can confirm that this was _not_ a problem in 2.3.

I suspect there's a problem with the introduction of weighting in the index vector. All of the following is from our test server:

Here's the metabib.title_field_entry row for "Mens creatix: an essay":

   id | source | field | value | index_vector
---------+--------+-------+------------------------+--------------------------------------------------------------
 3255651 | 83837 | 6 | mens creatrix an essay | 'an':3A,7C 'creatrix':2A,6C 'essay':4A,8C 'men':5C 'mens':1A
(1 row)

Note that "mens" gets stemmed to "men" in the weighting "C" but appears only as "mens" in weighting A. For this example, searches for both "men creatrix" and "mens creatrix" work, but "men's creatrix" returns no results.

The text search configuration, given "Men's creatrix", stems "Men's" down to "men" (and adds a separate word "s"):

SELECT ts_debug('english_nostop', 'Men''s creatrix');
                                     ts_debug
-----------------------------------------------------------------------------------
 (asciiword,"Word, all ASCII",Men,{english_nostop},english_nostop,{men})
 (blank,"Space symbols",',{},,)
 (asciiword,"Word, all ASCII",s,{english_nostop},english_nostop,{s})
 (blank,"Space symbols"," ",{},,)
 (asciiword,"Word, all ASCII",creatrix,{english_nostop},english_nostop,{creatrix})
(5 rows)

Weighting was new in 2.4, and with the rollback of part of the search changes, I'm wondering whether that affected the approach.

Alternately, perhaps QP is mangling the query such that the english_nostop text search configuration is not getting run against the incoming query?