On Mon, 25 Nov 2002, Eric S. Raymond wrote:
Michael Abbott <michael(_at_)araneidae(_dot_)co(_dot_)uk>:
In truth, a full solution would involve fetchmail hanging onto each
undeliverable mail and retrying the transmission in its own time. This
approach obviously has undesireable implications, but seems to me the only
complete solution.
That's not going to happen. The current "straight path" mode of operation is
important to the design; changing it might solve some edge cases but would
introduce way too many failure modes.
It's likely I may have misunderstood something, in particular I was
assuming the POP3 interface didn't allow *individual* messages to be
deleted. Watching further logs seems to show that I was wrong in assuming
this.
Instead, as far as I can tell, the problem I raised occurred because the
POP3 connection *timed out* while waiting for fetchmail to complete a
posting to sendmail. This presumably would have cancelled the message
deletions already performed, producing the behaviour I observed.
This problem has not recurred yet (presumably there was something
peculiarly intractable about the message that provoked the problem?) When
it does I'll dig a little deeper.