Re: [PATCH] Partial indices almost there

Martijn van Oosterhout <kleptog(at)svana(dot)org> writes:
> Does somewhere check for duplicate oids?

The duplicate_oids script in the same directory.

In practice, what really matters is that no two entries in the same
system catalog have the same OID; it wouldn't matter if we had, say,
the same OID used by a pg_proc and a pg_am entry. In some catalogs
it doesn't matter at all because we never do lookups by OID (this is
true for pg_attribute, for example *). We have a unique index on OID
on every system catalog where it matters, and ultimately that is what
ensures validity of the OID assignments. Keeping the hand-assigned
OIDs unique across the whole system is just a matter of neatnik-ism.

regards, tom lane

* In fact, Patricia Holben noticed awhile back that there *are*
duplicate OIDs in pg_attribute, in all releases <= 7.1. This was caused
by a bug in the code that creates pg_attribute rows for multi-key
indexes: it put the same OID on each such row. This is fixed in current
sources, but it's graphic proof that nonunique OIDs don't hurt when
they're not being used for lookup...