Re: BUG #3494: may be Query Error: subplan does not executed

Tom Lane wrote:
> It appears that join_clause_is_redundant() is rejecting the clause as
> redundant. I suppose some part of that machinery gets confused by the
> fact that the RHS of the clause references both relations. The
> EquivalenceClass rewrite cleaned this whole area up greatly, so no
> surprise that the bug is gone in HEAD. No time to look at it more now.

Yeah, that's it.

I don't understand that code very well. Why does it think that the right
pathkeys of "test1.id = test2.id" and "test1.id = test1.id+test2.id" are
equal?

Checking for overlapping left and right relation sets in
join_clause_is_redundant seems to fix the problem, though I'm not sure
if there's some other cases that that doesn't catch.

--
Heikki Linnakangas
EnterpriseDB http://www.enterprisedb.com