> > Note that elog(ERROR/FATAL) is changed to elog(STOP) if Critical
> > SectionCount > 0.
>
> Not in current sources ;-).
>
> Perhaps Vadim will say that I broke his error scheme, but if so it's
> his own fault for not documenting such delicate code at all.
Ok, it's my fault (though I placed NO ELOG(ERROR) comments everywhere
around critical xlog-related sections of code).
> I believe he's out of town this weekend, so let's wait till he gets
> back and then discuss it some more. Perhaps there is a need to
> distinguish xlog-related critical sections from other ones, or
> perhaps not.
Perhaps. For xlog-related code rule is simple - backend must not be
interrupted till changes are logged.
Vadim