Re: pg_dump versus views and opclasses

"Brendan Jurd" <direvus@gmail.com> writes:
> I recently had pg_dump produce a non-restorable dump for one of my
> databases.  I can't share the dump itself, but I can describe what
> went wrong. ...
>  * It seems there's no pg_depend entry for
> types/functions/operators/opclasses that the view depends on, unless
> they are part of the SELECT list.

What PG version exactly?  We've been moving towards fuller
representation of the semantics in the parse tree over time,
so that's a very relevant question.

FWIW I think this should be pretty much fixed as of CVS HEAD, because
all of the sorting/grouping semantics are now normalized in
SortGroupClauses and find_expr_references() does know about them.
Can you extract a test case from your problem DB so we can verify
nothing got missed?
        regards, tom lane