Re: Vote totals for SET in aborted transaction

Tom Lane wrote:
> Lincoln Yeoh  writes:
> > Coz some things should not be rolled back. So you guys might come up with a 
> > different keyword for it.
> 
> > CONFIG: for non transactional stuff that can appear as SQL statements.
> > SET: for stuff that can be transactional.
> 
> People keep suggesting this, and I keep asking for a concrete example
> where non-rollback is needed, and I keep not getting one.  I can't see
> the value of investing work in creating an alternative behavior when
> we have no solid example to justify it.
> 
> The "Oracle compatibility" argument would have some weight if we were
> making any concerted effort to be Oracle-compatible across the board;
> but I have not detected any enthusiasm for that.  Given that it's not
> even the same syntax ("SET ..." vs "ALTER SESSION ...") I'm not sure
> why an Oracle user would expect it to behave exactly the same.

Agreed.  OK, let me summarize.

We had a vote that was overwhemingly #1.  Marc made a good point that we
should see how other databases behave, and we now know that Oracle and
Ingres do #3 (honor all SETs in an aborted transaction).  Does anyone
want to change their vote from #1 to #3.

Second, there is the idea of doing #1, and having a GUC variable for #3.
Does anyone want that?  I think Marc may.  Anyone else?

--  Bruce Momjian                        |  http://candle.pha.pa.us pgman@candle.pha.pa.us               |  (610) 853-3000 +  If your life is a hard drive,     |  830 Blythe Avenue +  Christ can be your backup.        |  Drexel Hill, Pennsylvania 19026