With all this dependency stuff, what happens with the ALTER TABLE / DROP NOT
NULL syntax we came up with?
Maybe we should allow RESTRICT/CASCADE on that syntax and if restrict is
specified, you can't drop it if a primary key depends on it and if cascade
is specified it will drop the primary key...
Just for consistency...
Also, when talking about whether or not the index supporting a constraint
should be sort of 'hidden' from the user, should not we change pg_dump to
dump unique indices using the ALTER TABLE syntax, rather than the CREATE
UNIQUE INDEX syntax? Otherwise this information will be lost.
Chris