"Victor Y. Yegorov" <viy@mits.lv> writes:
> If I have on-disk bitmap
> ON (a, b, c)
> will the planner pick an index scan for
> WHERE a = 42 AND b = 'foo'
> (i.e. only part of the index attributes are involved)? Any modifications
> needed to achieve this functionality?
Hmm. That particular case will work, but the planner believes that only
consecutive columns in the index are usable --- that is, if you have
quals for a and c but not for b, it will think that the condition for c
isn't usable with the index. This is true for btree and gist indexes,
so I suppose we'd need to introduce a pg_am column that tells what to
do.
[ thinks some more ... ]
Plan B would be to remove that restriction and teach btree and gist to
cope. While a btree couldn't use a nonconsecutive restriction as part
of its where-to-scan logic, I don't see any good reason why it couldn't
still perform the test before returning the TID, thus possibly saving a
trip to the heap. Offhand it seems this should be true of gist as well,
but I don't know that code well enough to be sure.
regards, tom lane
