Tom,
* Tom Lane (tgl@sss.pgh.pa.us) wrote:
> Stephen Frost <sfrost@snowman.net> writes:
> > * Tom Lane (tgl@sss.pgh.pa.us) wrote:
> >> In further testing of that, I noticed that it made the behavior of our
> >> other bugaboo, the public schema, rather inconsistent. With this
> >> builtin-extensions hack, the plpgsql extension doesn't get dumped,
> >> whether or not you say "clean". But the public schema does get
> >> dropped and recreated if you say "clean". That's not helpful for
> >> non-superuser users of pg_dump, so I think we should try to fix it.
>
> > I'm not entirely sure about trying to also support --clean for
> > non-superusers.. We've long had that the public schema is dropped and
> > recreated with --clean and it seems likely that at least some users are
> > depending on us doing that. In any case, it's certainly not a change
> > that I think we could backpatch. Perhaps we could just change it moving
> > forward (which would make me happier, really, since what I think we do
> > with these initdb-time things currently is a bit bizarre).
>
> Sure, I was not proposing this for back-patch --- it depends on the
> other stuff we've committed recently, anyway.
>
> > Yes, having that in getNamespaces() isn't correct but we need to do
> > something there, and I've been trying to figure out what.
>
> I claim this is what ;-)
Well, that would make things in v11 better, certainly. I suppose for
back-patching, I can just go make the change to pg_restore that I
outlined and that would deal with the complaints we've gotten there.
I'm afraid we may still get some push-back from existing users of
--clean since, with the change you're proposing, we wouldn't be cleaning
up anything that's been done to the public schema when it comes to
comment changes or ACL changes, right?
A typical use-case of pg_dump with --clean being to 'reset' a system
after, say, dumping out some subset of a production system and then
loading it into the development environment that all of the devs have
full access to, and where there might have been changes to the 'public'
that you want to revert (to get it back to looking like how 'prod' was).
In current versions this should result in at least the ACLs on public
matching what they are on prod, along with the comment and any other
changes done to it, but we would lose that if we stop doing drop/create
of the public schema on --clean.
Then again, the DROP SCHEMA will fail if any objects end up created in
the public schema anyway, so it isn't like that's a complete solution
regardless. I suppose it's a bit surprising that we haven't been asked
for a --clean-cascade option.
Thanks!
Stephen
