Rod Taylor <rbt(at)rbt(dot)ca> writes:
> Ok.. If you think thats safe, I'll give it a try. I was afraid that the
> system would confuse itself if the table had mix and matched tuples in
> it. New tuples without oids, old tuples with.
Manfred's original implementation would have failed (since it didn't
have a tuple-header hasoid bit). I think I got all the places that
should consult the header bit, but there may be some left; you'll need
to test.
> That helps DROP OID. How about ADD OID?
What about it? I think it'll work just like adding a column, except
that OID will probably read as 0 not NULL if the row hasn't been updated
yet. (You could probably make it read as NULL if you wanted though.)
regards, tom lane