Greg Spiegelberg <gspiegelberg@cranel.com> writes:
> Tom Lane wrote:
>> Well, put a unique index on the contents column. Better to fail an
>> insert than to get a conflict of LO OIDs.
> Can't do that. Our app won't handle it.
Actually, I think the lo_import() will fail anyway, whether you like it
or not. There's a unique index on pg_largeobject.
> This does lead me to 2 questions... first, why is this still an issue and
> not fixed in the backend where OID's are managed?
When you're two major versions behind, you don't have a lot of leeway to
complain about why things are still an issue ;-). But the short answer
is that making OIDs 8 bytes would permanently break platforms that don't
have int64 support, and be a nontrivial performance hit on those where
int64 is substantially slower than int32. So I'd say it's a good ways
into the future yet. Eventually we'll decide we don't care about 32-bit
machines anymore, but not for awhile.
regards, tom lane