Tom Lane writes:
> What about requiring ownership of at least one of the types?
Yes, that would work.
There would be a somewhat bizzare consequence, though: User U1 creates
type T1, user U2 creates type T2. Then user U1 creates a cast from T1 to
T2. Now user U2 would be allowed to drop that cast (unless we store a
cast owner). I guess that lies in the nature of things.
A much more complex yet powerful alternative would be to associate casts
with schemas. For example, this would allow an ordinary user to create a
cast from numeric to text in his own little world. But that might be
going too far.
> > I'm not sure about the implications of associating objects with schemas in
> > pg_dump. I suppose there might be an option to dump only certain schemas,
>
> That is the intention (it's not implemented yet).
My concern was that if you, say, have two schemas and a cast that involves
types from both schemas. If you dump all of them, you have a consistent
dump. But if you dump both schemas separately, do you dump the cast in
both of them (thus making each schema's dump self-contained) or in only
one of them (thus allowing concatenation the dumps). This issue
generalizes to every kind of dependency in pg_dump.
--
Peter Eisentraut peter_e@gmx.net