Joe Conway <mail@joeconway.com> writes:
> I've been seeing 2 regression failures (diff attached) for the past
> couple of days. Both appear to be ordering related.
Yeah, I'm getting that too. It seems to be a side effect of my "fuzzy
cost comparison" patch. I've been trying to figure out why I did not
notice it before committing. I don't normally make commits without
checking the regression results --- did I miss a step, or is there
something else involved? Is there anyone who is using CVS from the last
day or so and does *not* see these diffs?
Anyway, the new choices of plans appear to be reasonable AFAICT,
so I will probably end up just changing the regression expected
outputs. In the select_views case, for instance, it is now preferring
a plan with cost 1.05 .. 709.74-plus-epsilon over one with cost
132.24 .. 709.74 (the total costs are now considered equal so it looks
to the startup cost as tiebreaker). Seems like a win to me.
regards, tom lane