Richard Guo <guofenglinux@gmail.com> writes:
> Hmm. Doing this complicates matters even more. Maybe we can just loosen
> the cross-check for nullingrels to cope with this change by using
> NRM_SUBSET matches for joinquals (including mergeclauses, hashclauses
> and hashkeys) in set_join_references()?
That would be sad ... it'd basically be conceding defeat at the task
of knowing that we've accurately placed joinquals, which is one of
the most fundamental things I wanted to get out of this rewrite.
I might accept weakening those assertions as a stopgap that we plan to
work on more later, except that I'm afraid that this is telling us
there are still bugs in the area.
What's feeling like it might be the best thing is to go ahead and
syntactically convert to the second form of identity 3 as soon as
we've determined it's applicable, so that upper C Vars are always
marked with both OJ relids. Not sure how much work is involved
there.
regards, tom lane