Peter Eisentraut <peter.eisentr...@enterprisedb.com> writes: > Several programs wrap, say, pg_log_fatal() into a pg_fatal(), that does > logging, cleanup, and exit, as the case may be. I think that's a good > solution.
I agree, and my draft patch formalized that by turning pg_log_fatal into exactly that. The question that I think is relevant here is what is the point of labeling errors as "error:" or "fatal:" if we're not going to make any serious attempt to make that distinction meaningful. I'm not really buying your argument that it's fine as-is. Anybody who thinks that there's a difference is going to be very confused by the behavior they observe. But, if we all know there's no difference, why have the difference? regards, tom lane