strk(at)refractions(dot)net writes:
> Doing some tests I've found out that the returned value from the
> JOINSEL is applied to REL1.rows X REL2.rows, but REL1 and REL2
> are not 'base' table, rather relations with a number of
> rows once again estimated by other selectivity functions.
Right. This amounts to assuming that the join conditions and the
restriction conditions are independent, which of course is bogus,
but we really don't have enough information to do better.
regards, tom lane