Alexey Bashtanov <bashtanov@imap.cc> writes:
> This is how one can create a parent with a non-nullable column and a
> child with the same column nullable:
Yeah, that's been on the to-do list for a long time. The current
theory about it is that NOT NULL constraints ought to be represented
as entries in pg_constraint like CHECKs are, so that the inheritance
count and so forth could be managed the same way. We could keep
attnotnull for convenience but it'd just mirror the existence of
the pg_constraint entry.
IIRC Alvaro had a WIP patch for this, but I've not heard anything
about it lately.
regards, tom lane