Martijn van Oosterhout <kleptog@svana.org> writes:
> On Mon, Apr 10, 2006 at 04:40:50PM +0800, Qingqing Zhou wrote:
>> AFAIK if two rows are with the same OID, we will think these two rows are
>> different versions of the same tuple. So if we allow users to change OID
>> columns, we may encounter some inconsistency.
> Hardly, OIDs are optional anyway these days. We only guarentee that
> OIDs will be unique if you have a unique index on them, so I imagine
> the only reason it hasn't happened yet is because nobody's coded it.
Right. I looked at this awhile back, though, and found multiple places
that would break if you tried it :-( --- mainly assumptions about the
meaning of resno in target lists. (The comments for the TargetEntry
struct give some but not all of the details.) Given that OID is
semi-deprecated anyway these days, it's not clear it's worth fixing.
regards, tom lane