On Sat, Aug 12, 2023 at 12:56 AM Andres Freund <andres@anarazel.de> wrote:
> On 2023-08-11 11:56:27 -0700, Peter Geoghegan wrote:
> > On Fri, Aug 11, 2023 at 11:23 AM Andres Freund <andres@anarazel.de> wrote:
> > > > Couldn't you say the same thing about defensive "can't happen" ERRORs?
> > > > They are essentially a form of assertion that isn't limited to
> > > > assert-enabled builds.
> > >
> > > Yes. A lot of them I hate them with the passion of a thousand suns ;). "Oh,
> > > our transaction state machinery is confused. Yes, let's just continue going
> > > through the same machinery again, that'll resolve it.".
> >
> > I am not unsympathetic to Ashutosh's point about conventional ERRORs
> > being easier to deal with when debugging your own code, during initial
> > development work.
>
> Oh, I am as well - I just don't think it's a good idea to introduce "log + error"
> assertions to core postgres, because it seems very likely that they'll end up
> getting used a lot.
>
>
I am open to ideas which allow the same backend to recover after
meeting an easily recoverable but "can't happen" condition rather than
losing that backend and starting all over with a new backend. Not all
Assert'ed conditions are recoverable so a blanket GUC or compile time
option won't help. Those might make things worse. We need two separate
incantations for non-recoverable and recoverable Asserts respectively.
I like Peter's idea of having a new elevel, however it still requires
adding conditional USE_ASSERT, an if testing the condition and then
writing an error message. AssertLog() in the patch uses just a few
more letters.
It won't help to expand the scope of the problem since that will
reduce the chances of getting anything done.
--
Best Wishes,
Ashutosh Bapat
