Jim Nasby <Jim.Nasby@BlueTreble.com> writes:
> On 2/14/17 1:18 PM, Tom Lane wrote:
>> One point that could use further review is whether the de-duplication
>> algorithm is actually correct. I'm only about 95% convinced by the
>> argument I wrote in planunionor.c's header comment.
> I'll put some thought into it and see if I can find any holes. Are you
> only worried about the removal of "useless" rels or is there more?
Well, the key point is whether it's really OK to de-dup on the basis
of only the CTIDs that are not eliminated in any UNION arm. I was
feeling fairly good about that until I thought of the full-join-to-
left-join-to-no-join conversion issue mentioned in the comment.
Now I'm wondering if there are other holes; or maybe I'm wrong about
that one and it's not necessary to be afraid of full joins.
regards, tom lane