Christopher Kings-Lynne <chriskl@familyhealth.com.au> writes:
>> I don't think we have ever changed oids for existing data types, so you
>> should be OK.
> Are you sure? If we remove a type, then its oid becomes up for grabs by
> the unused_oids script.
But if we remove a type then it's a bit moot whether the OID for it
stays constant. For types that have remained present across all
versions I don't think we have ever changed the OID assignment.
I'd be willing to promise that this will stay true for the types
that are "manually" declared in pg_type.h. (As a counterexample,
I think that information_schema.sql creates some domain types, and
those types aren't going to have frozen OIDs.)
> I think the core should mandate it as policy never to reuse oids and
> perhaps make the unused_oids script "remember" what has been used...
Absolutely not worth the trouble. If we were to guarantee not to remove
types, then there might be some value in making such a promise, but we
haven't and won't. (Precedent: the SQL spec itself has added and
dropped datatypes.)
regards, tom lane
