"Inoue, Hiroshi" <h-inoue@dream.email.ne.jp> writes: > Could you please tell the customer that it's meaningless to set > client_min_messages to 'fatal' or 'panic'?
Yeah, I wonder why we allow that at all. It's basically breaking the wire protocol ...
Agreed. I'll send in a patch if you want. It appears the fatal and panic members of client_message_level_options in guc.c could be removed. The same options are similarly used by trace_recovery_messages, but that seems to make sense even in that regard. Likewise, postgresql.conf.sample only shows debug5-error as options; omitting fatal and panic. The web docs, however, show fatal and panic as options.