Re: Incorrect handling of IS [NOT] NULL quals on inheritance parents

Hmm, I have thought about your point, but I may have a different
perspective.  I think the oversight discussed here occurred because we
mistakenly recorded NOT NULL columns that are actually nullable for
traditional inheritance parents.  Take the query from my first email as
an example.  There are three RTEs: p(inh), p(non-inh) and c(non-inh).
And we've added a NOT NULL constraint on the column a of 'p' but not of
'c'.  So it seems to me that while we can mark column a of p(non-inh) as
non-nullable, we cannot mark column a of p(inh) as non-nullable, because
there might be NULL values in 'c' and that makes column a of p(inh)
nullable.

And I think recording NOT NULL columns for traditional inheritance
parents can be error-prone for some future optimization where we look
at an inheritance parent's notnullattnums and make decisions based on
the assumption that the included columns are non-nullable.  The issue
discussed here serves as an example of this potential problem.

Thanks
Richard