Jan Wieck <JanWieck@yahoo.com> writes:
> IMHO there's nothing fundamentally wrong with having pg_dump
> dumping the constraints as special triggers, because they are
> implemented in PostgreSQL as triggers. ...
> The advantage of having pg_dump output these constraints as
> proper ALTER TABLE commands would only be readability and
> easier portability (from PG to another RDBMS).
More to the point, it would allow easier porting to future Postgres
releases that might implement constraints differently. So I agree with
Philip that it's important to have these constructs dumped symbolically
wherever possible.
However, if that's not likely to happen right away, I think a quick hack
to restore tgconstrrelid in the context of the existing approach would
be a good idea.
regards, tom lane