Nathaniel writes...
The problem is that knowing what is "the recipient's mailbox" is pretty much
impossible in the general case. For example, when UNIX sendmail writes
something into /usr/spool/mail/username, it thinks of that as the
recipient's mailbox, and the case is closed. However, I know of cases in
which a gateway to another mail world is implemented as something that
checks /usr/spool/mail files and picks things up from there, in which view
/usr/spool/mail is just an intermediate queueing area. In general, MTA's
simply don't know whether their "delivery" of mail is final or intermediate,
which is why I think this kind of ack isn't useful even in the case you
describe. You can even get a positive "it was absolutely delivered" ack
from sendmail in such a case, and then get back a REJECTION from something
further downstream.
Neither kind of acknowledgement mechanism will handle this case correctly.
When a mail message is forwarded after delivery, it should have the envelope
return address set to the forwarding agent, not the sender of the original
message.
The problem here is once again that message headers are being used to affect
transport. Another way to state the problem is that the use of certain
headers like Sender: and Return-path: in conjunction with mail transport is
not well specified. This needs to be addressed anyway, regardless of what
acknowledgement mechanisms we propose.
Keith