Tom Lane wrote:
> What I'm thinking of doing is removing the code, and replacing the
> README with a note explaining how to convert contrib/array queries to
> use the new mainstream syntaxes. That will give contrib/array users
> a clue what they're supposed to do. In a release or three the README
> could go away too.
I have no objection to removing it now, but previously I think you
agreed with Bruce's comment that we should leave it intact (but
deprecated) for 7.4, and remove in 7.5.
> Also, does anyone want to look for possible dead code in intagg and
> so on?
I did a quick review back in July.
IIRC, intagg could be functionally replaced with the aggregate
definition that I posted, except that intagg is probably a fair bit
better performance (I didn't actually test), in that it accumulates the
array in backend memory and just pushes pointers around as int4's. I've
thought that a safer implementation would be needed to fold it into the
backend (maybe using hashes keyed with the pointer?), but in any case
that's a 7.5 thing.
Also IIRC there were some functions in intarray that overlap the new
backend functionality, but much of it (i.e. using GIST to index into
large arrays) is not.
I'll try to review them again and make a recommendation in the next
couple of days, but it might be a stretch because I'm trying to tie up
lots of loose ends in preparation for a trip next week.
Joe