I noticed a couple revs ago (haven't tried yet today though) that reopening stderr as /var/log/dbmail/dbmail.err was happening a) before the config file was read, and b) after stderr was already closed. This means that if the daemon can't find the config file, or the logging directory does not exist, it bombs without any detectable error. That's not so good.
Paul, have you had a chance to test these situations? The most important thing here is that we don't lose access to the "real stderr" if the new stderr file can't be created. Aaron
