On Wed, 31 Jan 2024 at 18:23, Robert Haas <robertmhaas@gmail.com> wrote:
> That's kind of an odd artifact, but maybe it's fine in
> practice.
I agree it's an odd artifact, but it's not a regression over the
status quo. Achieving that was the intent of my suggestion: A change
that improves some cases, but regresses nowhere.
> I say again that it's good to test out a bunch of scenarios
> and see what shakes out.
Testing a bunch of scenarios to find a good one sounds like a good
idea, which can probably give us a more optimal heuristic. But it also
sounds like a lot of work, and probably results in a lot of
discussion. That extra effort might mean that we're not going to
commit any change for PG17 (or even at all). If so, then I'd rather
have a modest improvement from my refinement of Melih's proposal, than
none at all.