On Thu, 19 Nov 1998, Lars Kellogg-Stedman wrote:
> I was going to ask if it might make sense for exim to go the sendmail route
> and not make any attempt at handling local deliveries (you know, move the
> filtering language out into a seperate binary, maybe have a small,
> lightweight delivery program if /bin/mail isn't small and lightweight enough).
>
> I'm not sure if this would have any impact on exim's performance as an
> mta. Obviously, it would mean a little more work for local deliveries, but
> I can't guess how much.
Several points here.
Firstly, you can already configure Exim not to do local deliveries if
you don't want to. There are sites that do everything via procmail.
You don't even have to have the appendfile code in the binary.
Secondly, Exim's filtering has different effects to what you would get
if you did it in a separate process that the message was piped to, like
procmail. For example, if a separate process wants to forward a file to
another address, it has to resubmit the message to the MTA which then
treats it as an entirely separate message. All it sees is a user process
submitting a new message - it can't tie it up with the message which is
forwarded. Maybe this is OK, and maybe it is even desirable, but the
point I'm making is that it is a different way of doing things. It would
certain make the tracing of forwarded messages by the mailadmin harder.
Thirdly, Exim's filtering stuff does provide the system filter
capability as well as user filtering. This seems to be quite popular.
Exim did not originally have any filtering facilities. I was prevailed
upon to create them because, it was argued, it is convenient to have
them in the MTA for efficiency, and also because the MTA can do better
in cases of error. Consider: A message comes in from sender A addressed
to user B, who forwards it to address C. However, the attempt to deliver
to C bounces for some reason. Using Exim's filtering, the bounce message
will go to user A. However, if Exim has "delivered" the message to user
B's local delivery process which has reinjected it, addressed to user C, then
the bounce message will come back to B, which isn't helpful. This of
course is assuming that the user process is unprivileged. If it is
run under some special uid that Exim trusts, then this particular
problem can be circumvented if everything is set up correctly.
--
Philip Hazel University of Cambridge Computing Service,
ph10@??? Cambridge, England. Phone: +44 1223 334714.
--
*** Exim information can be found at
http://www.exim.org/ ***
