Alexey Bashtanov <bashtanov@imap.cc> writes:
> Is there a reason that join cardinality estimates are not limited by the
> product of the joined parts cardinalities like in the
> join-card-est.patch attached?
Because that would be giving an unfair advantage to some paths over
others based on nothing except estimation errors. I do not think we'd
get a net benefit in plan quality.
If we could do this earlier and adjust the join relation's overall
cardinality estimate, it might be something to consider.
regards, tom lane