Robert Haas <robertmhaas@gmail.com> writes:
> I think it would be over the top to suggest that pg_dump has to cope
> with modifications that can only occur through manual updates to the
> system catalogs, but it seems like anything that can be done using DDL
> statements should be handled.
Like, say, DELETE FROM pg_proc?
Basically, if you use superuser powers to fool with the definition of
any system object, it's on your own head whether the result works at all
and whether or how to preserve that change into a new version of
Postgres. I do not think it's part of pg_dump's charter to try to
handle that. In a significant fraction of cases, preserving the change
would be exactly the wrong thing, but there is no way for pg_dump to
know the difference.
regards, tom lane