On Fri, 28 Jul 2006, Mark Lewis wrote:
> Does the current savepoint implementation in the driver require a
> separate round-trip?
Yes. Creating, rolling back to, or releasing a savepoint each currently
make a server round-trip. If you're suggesting queueing these up so that
it sends these with queries instead of alone, I think that's something the
driver could do if it was creating savepoints itself, but would be tougher
to do for user savepoints. For user savepoints I think you need to get
feedback to the caller immediately upon command execution rather than with
a later query. Consider something like trying to release an invalid
savepoint. You need to tell the user immediately.
Kris Jurka