Robert Haas <robertmhaas@gmail.com> writes:
> On Wed, Jul 27, 2011 at 7:51 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> I think pg_dumpall is the very least of your problems if you do
>> something like that. �We probably ought to forbid it entirely.
> Well, we had a long discussion of that on the thread Phil linked to,
> and I don't think there was any consensus that forbidding it was the
> right thing to do.
You're right, I was half-remembering that thread and thinking that
there are a lot of gotchas in doing an ALTER ROLE SET ROLE. Florian
claimed in the thread that he'd never hit one before, but that doesn't
convince me much.
> Phil appears to be trying to implement the
> proposal you made here:
> http://archives.postgresql.org/pgsql-hackers/2011-01/msg00452.php
> ...although I don't think that what he did quite matches what you asked for.
No, the proposed patch doesn't go nearly far enough to address Florian's
problem. What I was speculating about was moving all the role (and
database) alters to the *end*, so they'd not take effect until after
we'd completed all the restore actions.
regards, tom lane