Re: Check SubPlan clause for nonnullable rels/Vars

Richard Guo  writes:
> [ v2-0001-Check-SubPlan-clause-for-nonnullable-rels-Vars.patch ]

Pushed with cosmetic changes:

* I don't believe in "add at the end" as a principle for placement
of new code.  There's usually some other logic that will give more
consistent results.  In cases like this, ordering the treatment of
Node types in the same way as they appear in the include/nodes/
headers is the standard answer.  (Not that everybody's been totally
consistent about that :-( ... but that's not an argument for
introducing even more entropy.)

* I rewrote the comments a bit.

* I didn't like the test case too much: spinning up a whole new set
of tables seems like a lot of useless cycles.  Plus it makes it
harder to experiment with the test query manually.  I usually like
to write such queries using the regression database's standard tables,
so I rewrote this example that way.

			regards, tom lane