At 01:20 AM 12/07/2007, Dmitry Yemanov wrote:

>Helen Borrie wrote:
> >
> > I *don't* think that's likely.

Dmitry,

>But we do support them implicitly already, i.e. for the ON CONNECT
>triggers. Personally, I do see value in autonomous transactions (e.g.
>for audit/logging purposes).

I concede that an autonomous transaction would be *useful* for
logging purposes. I just shudder at the idea of allowing the
autonomy principle to creep into the picture and allow Tom, Dick and
Harry to change database state from an uncommitted transaction.

I had the same kind of reservations about EXECUTE STATEMENT. At the
time, its proponents argued, "Of course people should know better
than to use it to execute DDL". Correct. Of course they
should. But it's a honey-pot for idiotic workarounds, and
hit-and-run transactions would be, too, with similar kinds of
untrackable corruptions that EXECUTE STATEMENT has made possible.

The idea of pragmatising idiot-prone features has merit, though, I
have to say. :-)

./heLen