On 2024-Feb-27, Michael Paquier wrote:
> These would cause compilation failures. Saying that, this is a very
> nice cleanup, so I've fixed these and applied the patch after checking
> that the one-one replacements were correct.
Oh, I thought we were going to get rid of ObjectClass altogether -- I
mean, have getObjectClass() return ObjectAddress->classId, and then
define the OCLASS values for each catalog OID [... tries to ...] But
this(*) doesn't work for two reasons:
1. some switches processing the OCLASS enum don't have "default:" cases.
This is so that the compiler tells us when we fail to add support for
some new object class (and it's been helpful). If we were to
2. all users of getObjectClass would have to include the catalog header
specific to every catalog it wants to handle; so tablecmds.c and
dependency.c would have to include almost all catalog includes, for
example.
What this says to me is that ObjectClass is/was a somewhat useful
abstraction layer on top of catalog definitions. I'm now not 100% that
poking this hole in the abstraction (by expanding use of catalog OIDs at
the expense of ObjectClass) was such a great idea. Maybe we want to
make ObjectClass *more* useful/encompassing rather than the opposite.
(*) I mean
Oid
getObjectClass(const ObjectAddress *object)
{
/* only pg_class entries can have nonzero objectSubId */
if (object->classId != RelationRelationId &&
object->objectSubId != 0)
elog(ERROR, "invalid non-zero objectSubId for object class %u",
object->classId);
return object->classId;
}
plus
#define OCLASS_CLASS RelationRelationId
#define OCLASS_PROC ProcedureRelationId
#define OCLASS_TYPE TypeRelationId
etc.
--
Álvaro Herrera Breisgau, Deutschland — https://www.EnterpriseDB.com/
Bob [Floyd] used to say that he was planning to get a Ph.D. by the "green
stamp method," namely by saving envelopes addressed to him as 'Dr. Floyd'.
After collecting 500 such letters, he mused, a university somewhere in
Arizona would probably grant him a degree. (Don Knuth)