Robert Haas <robertmhaas@gmail.com> writes:
> EDB has an implementation of this in Advanced Server. A stored
> procedure can issue a COMMIT, which commits the current transaction
> and begins a new one. This might or might not be what people are
> imagining for this feature. If we end up doing something else, one
> thing to consider is the impact on third-party tools like PGPOOL,
> which currently keep track of whether or not a transaction is in
> progress by snooping on the stream of SQL commands. If a procedure
> can be started with no transaction in progress and return with one
> open, or the other way around, that method will break horribly.
> That's not necessarily a reason not to do it, but I suspect we would
> want to add some kind of protocol-level information about the
> transaction state instead so that such tools could continue to work.
Huh? There's been a transaction state indicator in the protocol since
7.4 (see ReadyForQuery). It's not our problem if PGPOOL is still using
methods that were appropriate ten years ago.
regards, tom lane