Andres Freund <andres@anarazel.de> writes:
> On 2018-11-06 11:37:40 -0500, Tom Lane wrote:
>> Andres Freund <andres@anarazel.de> writes:
>>> Seems reasonable. I do think it's probably sensible to backpatch,
>>> although I wonder if we shouldn't clamp the value to ERROR at log
>>> emission error time, rather than via guc.c, so we don't prevent old code
>>> / postgresql.conf that set client_min_messages to > ERROR.
>> Hm, do you really think there is any?
> I'm not sure. But it sounds like it'd possibly slow adoption of the
> minor releases if we said "hey, make sure that you nowhere set
> client_min_messages > ERROR", even if it's not particularly meaningful
> thing to do, as it'd still imply a fair bit of work for bigger
> applications with not great standards.
OK, so the consensus seems to be that the back branches should continue
to allow you to set client_min_messages = FATAL/PANIC, but then ignore
that and act as though it were ERROR.
We could implement the clamp either in elog.c or in a GUC assignment
hook. If we do the latter, then SHOW and pg_settings would report the
effective value rather than what you set. That seems a bit cleaner
to me, and not without precedent. As far as the backwards compatibility
angle goes, you can invent scenarios in which either choice could be
argued to break something; but I think the most likely avenue for
trouble is if the visible setting doesn't match the actual behavior.
So I'm leaning to the assign-hook approach; comments?
regards, tom lane