Vaclav Juza <vaclav.juza@xitee.com> writes:
> I have modified the test case (it is more similar to the real-word
> query), so that now it is slower on 9.2.1 than on 9.1.4 (the version
> with the explicit cast):
FWIW, testing this on HEAD (with the patch I committed last week),
I get a plan that's about 10% faster than 9.1; or if I disable
index-only scans for a more level playing field, it's just about
equivalent to 9.1.
regards, tom lane