Tom Lane wrote:
> Alvaro Herrera <alvherre@commandprompt.com> writes:
>> I'm not sure we can use the simple "raise an ERROR" answer though,
>> because for users that would be a regression.
>
> I've reconsidered the idea of upgrading the outer xact's shared lock to
> exclusive: at first I thought that would be hard to implement correctly,
> but now I realize it's easy. Just re-use the XID that's in the multixact
> as the one to store as the exclusive locker, instead of storing our
> current subxact XID. In some cases this will be a subcommitted XID of
> the current subxact or a parent, but the locking semantics are the same,
> and even though we think such an XID is finished everyone else will see
> it as still live so the appearance of its XID in an XMAX field shouldn't
> be an issue.
That way, the lock won't be downgraded back to a shared lock on
"rollback to savepoint", right? Though it's still better than throwing
an error, I think.
-- Heikki Linnakangas EnterpriseDB http://www.enterprisedb.com