Joshua D. Drake wrote:
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
On Sun, 02 Mar 2008 15:46:25 -0600
Karl Denninger <karl@denninger.net> wrote:
I'm not quite clear what I have to do in terms of if/when I can drop
the old tsearch config stuff and for obvious reasons (like not
running into this in the future) I'd like to. Can I just run the
"untsearch2" script against those databases or will that destroy the
search functionality? Are there changes necessary in the SQL code
(the documentation implies not unless I'm doing "odd" things)
One of the goals for 8.3 and the integrated tsearch was to remove
exactly this problem.
Sincerely,
Joshua D. Drake
Unfortunately there are code changes necessary on my end in one of my major applications; the attempt at compatability is somewhat less than successful from here.....
The problem is that I was holding the ts_vector in a column in the table with a GIST index on that column. This fails horribly under 8.3; it appears to be ok on the reload but as there is a trigger on updates any update or insert fails immediately with a data mistype complaint.
The fix is to rejigger the query to go directly at the fields and build a gin or gist index directly on the underlying, which is not difficult but DOES require code and schema changes. I'll get through it but this is going to bite people with some authority if they have applications that were doing things the same way I was - there was nothing esoteric about the way I had coded it (although it could be argued it was somewhat wasteful of disk space.)
-- Karl Denninger
karl@denninger.net