Josh Berkus wrote:
> Bruce,
>
> > The ordering of the lexems was changed:
> >
> > /*
> > * Convert 8.3 tsvector to 8.4
> > *
> > * 8.3 sorts lexemes by its length and if lengths are the same then it uses
> > * alphabetic order; 8.4 sorts lexemes in lexicographical order, e.g.
> > *
> > * => SELECT 'c bb aaa'::tsvector;
> > * tsvector
> > * ----------------
> > * 'aaa' 'bb' 'c' -- 8.4
> > * 'c' 'bb' 'aaa' -- 8.3
> > */
>
> What does that get us in terms of performance etc.?
>
> Given that it's going to prevent users of Tsearch from
> upgrading-in-place, the format change ought to be giving us some serious
> gains. Otherwise we should put it off until we need to make other
> datatype changes.
No idea but now that we are in beta we would then be invalidating beta
tester data. The area I got stuck on is that there is no CAST behavior
when creating an index. We are already invalidating GIN indexes, but I
can't even create an index to support the old data type.
-- Bruce Momjian <bruce@momjian.us> http://momjian.us EnterpriseDB
http://enterprisedb.com
+ If your life is a hard drive, Christ can be your backup. +