Manfred Koizar <mkoi-pg@aon.at> writes:
> Revised version of the first in a series of patches leading to
> heap tuple headers withoid oid for tables created WITHOUT OIDS.
My objection to this remains the same as before: heap_getsysattr
now has a core-dump condition built into it, because it can't tell
whether it's safe to do HeapTupleGetOid().
You really need to propagate relhasoids into tuple descriptors,
which is what heap_getsysattr could be passed from heap_getattr.
Having done that, most if not all of the AssertRelationHasOids()
things should be replaced by tests on tuple descriptors; you
might also find a few more places where it's possible to make
the assertion test, rather than having to write
> + /* no Relation here to AssertRelationHasOids(relation); */
regards, tom lane