Tom Lane wrote:
Anecdotally, I had a situation recently where I got different plans
depending on whether I queried a join of a view against itself, or
"macro expanded" the view by hand. I was =very= surprised at this.
Me too, at least if you didn't do any hand optimization but just stuck
the view definition in as a sub-select. Can you provide a reproducible
case?
No surprise, I cannot reproduce this, especially since the DDL has
evolved since then. But you are almost certainly right, my "macro
expansion" must have done something more than simply dropping in the
view definition as is. Presumably something minor (to me) but a
show-stopper for the planner.
Sorry for the distraction. :(
- John D. Burger
MITRE