Re: New design for FK-based join selectivity estimation

Hi,

Attached is a reworked patch, mostly following the new design proposal
from this thread.

I'm not entirely happy with the code, but it's the best I've been able
to come up with by now and I won't be able to significantly improve that
due to travel over. Inevitably there will be issues in the code, and if
there's a chance of fixing them I'll do my best to do that over the
evenings at a hotel.

The filtering and matching to eclasses / join quals is triggered from
planmain.c - I believe this is the right place and that those pieces are
reasonably solid.

The estimation still happens in costsize.c, of course, but was modified
to use the pre-matched info. I have my doubts about this part, and I'm
sure Tom had something less complex / more efficient in mind (using the
pre-matched info).

regards

--
Tomas Vondra http://www.2ndQuadrant.com
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services