Greg Stark <gsstark@gmail.com> writes:
> A lot of work has gone into making pg_dump/pg_restore guarantee that
> they'll always produce a copy of the database, even if you've done odd
> things like change the lower bounds of your arrays. A lot of this was
> from before the days of PITR when pg_dump/pg_restore was the *only*
> backup option and it was considered absolutely essential that they
> always work. But even today I think it's still a goal that pg_dump
> always dump a loadable database.
Well, pg_upgrade still depends on pg_dump being a 100% solution for DDL,
so I don't think the requirements have gone down any ...
> I had in mind for pg_dump to decide to use the non-standard syntax iff
> it was necessary at dump time.
Maybe. I'm concerned about the cost of determining whether it's
necessary ("cost" meaning both "runtime" and "code complexity").
regards, tom lane