David Rowley <dgrowleyml@gmail.com> writes:
> On Wed, 9 Apr 2025 at 03:32, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> The simplest fix I can think of is to disable the presorted-agg
>> optimization if (1) there's a FILTER clause and (2) the proposed
>> sort key is anything more complex than a Var.
> Unfortunately, the situation is a little worse than what you
> highlighted, as I think I didn't consider FILTER at all, and this
> means I didn't consider the costing differences between filtering then
> sorting vs sorting then filtering.
Oooh. If the FILTER clause is selective, that could easily mean that
the "optimization" loses big from having to sort many more tuples.
I wonder if we should just not apply it when there's a FILTER,
full stop.
regards, tom lane
