Nikhil Sontakke <nikkhils@gmail.com> writes:
>> (Also, don't forget you need to hack pg_dump, too.)
> Yeah, I have already hacked it a bit. This constraint now needs to be
> spit out later as an ALTER command with ONLY attached to it
> appropriately. Earlier all CHECK constraints were generally emitted as
> part of the table definition itself.
IIRC, there's already support for splitting out a constraint that way,
in order to deal with circular dependencies. You just need to treat
this as an additional reason for splitting.
regards, tom lane