Richard Guo <guofenglinux@gmail.com> writes:
> On Thu, Aug 29, 2024 at 4:47 AM Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> In the meantime, I think this test case is mighty artificial,
>> and it wouldn't bother me any to just take it out again for the
>> time being.

> Yeah, I think we can remove the 't1.two+t2.two' test case if we go
> with your proposed patch to address this performance regression.

Here's a polished-up patchset for that.  I made the memoize test
removal a separate patch because (a) it only applies to master
and (b) it seems worth calling out as something we might be able
to revert later.

I found one bug in the draft patch: add_nulling_relids only processes
Vars of level zero, so we have to apply it before not after adjusting
the Vars' levelsup.  An alternative could be to add a levelsup
parameter to add_nulling_relids, but that seemed like unnecessary
complication.

                        regards, tom lane