There's a thread on the ODBC list[1] complaining about the fact that
it's possible to set client_min_messages to FATAL or PANIC, because
that makes ODBC misbehave. This is not terribly surprising, because
doing so arguably breaks the frontend protocol. The simple-query
section says this:
In the event of an error, ErrorResponse is issued followed by
ReadyForQuery.
and the extended-query section says this:
Therefore, an Execute phase is always terminated by the appearance of
exactly one of these messages: CommandComplete, EmptyQueryResponse (if
the portal was created from an empty query string), ErrorResponse, or
PortalSuspended.
and both of those are lies if an ERROR response gets suppressed thanks to
client_min_messages being set too high. It seems that libpq+psql manages
to survive the case (probably because psql is too stupid to notice that
anything is wrong), but I don't find it unreasonable that other clients
get hopelessly confused.
Hence, I propose that we should disallow setting client_min_messages
any higher than ERROR, and that this probably even amounts to a
back-patchable bug fix.
Thoughts?
regards, tom lane
[1] https://www.postgresql.org/message-id/flat/EE586BE92A4AFB45B03310C2A0C0565D6D0EFC17%40G01JPEXMBKW03