Chris Dunlop <chris@onthe.net.au> writes:
> E.g. using the script below, the 'bar.f1' column in the 'new'
> database ends up with a 'not null' constraint that isn't present
> in the 'orig' database.
> create table foo (f1 integer not null);
> create table bar () inherits(foo);
> alter table bar alter column f1 drop not null;
The general consensus is that the above should be illegal, ie, the ALTER
should have been rejected. Otherwise you would have a situation where a
"SELECT FROM foo" could return nulls, violating the very clear contract
of that table. We have not got around to enforcing this yet, but it's
on the TODO. I don't see it as a pg_dump bug that it's unable to
reproduce an invalid situation.
regards, tom lane