does ispell have allaffixes set to on?

[Brian <brian@xxxxxxxxxxxxxxxxxxxx>]

I was testing the ispell text search dictionary and it appears to be behaving as 
if the ispell option "allaffixes" was set to "on".  This wasn't the case for the 
original tsearch2 contrib module, and for the ispell program itself which 
defaults to "off".

So for example, if I create a simple DictFile with an entry for the word "brand" 
 (brand/DGRS) and a simple english affix AffFile that does those standard 
ispell suffixes (*D > ED, *G > ING, *R > ER and *S > S) along with the standard 
ispell prefixes (*A: . > RE, *I: . > IN, *U . > UN) then the ispell dictionary 
will return a lexeme for any input token containing a suffix and one of those 
prefixes even though NONE of the prefixes have been listed in the dictionary 
file as active for that word.

The following is observed and expected:

mydb=> CREATE TEXT SEARCH DICTIONARY test_ispell (
    TEMPLATE  = ispell,
    DictFile  = test,
    AffFile   = test,
    StopWords = english );

mydb=> SELECT
    ts_lexize('test_ispell', 'branding')  AS sfx_yes,
    ts_lexize('test_ispell', 'brandest')  AS sfx_no,
    ts_lexize('test_ispell', 'notindict') AS dict_no,
    ts_lexize('test_ispell', 'rebrand')   AS pfx_no;
 sfx_yes | sfx_no | dict_no | pfx_no
---------+--------+---------+--------
 {brand} |        |         |
(1 row)


However, the following results are NOT expected:

mydb=> SELECT
    ts_lexize('test_ispell', 'unbranded')  AS sfx_wpfx1,
    ts_lexize('test_ispell', 'rebranding') AS sfx_wpfx2;
 sfx_wpfx1 | sfx_wpfx2
-----------+-----------
 {brand}   | {brand}
(1 row)

In that second statement I expect NULL values indicating that the tokens are 
unknown, rather than lexemes indicating a match.  Is this expected behavior or a 
bug, and is there any way to control this?  Before I try to patch this in the 
code I'd like to know if it's intentional behavior or not.

It gets even screwier if you add "rebrand" to the dictionary (e.g. rebrand/DGS).
Then ts_lexize('test_ispell', 'rebranding') returns an array of both lexemes 
"{rebrand,brand}", when only the first is anticipated and wanted.

Thanks,

Brian Carp

--
Sent via pgsql-general mailing list (pgsql-general@xxxxxxxxxxxxxx)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-general