Nathaniel, et al,
I concur that the time has come to consider such things as you suggest.
And I believe that keeping the scope manageable is required.
But the issue of acks brings us into the more general question of
viewing email as a kind of packet network and, therefore, looking to
using other packet-net solutions for other email problems. Further, I
think that we should try very hard to distinguish UA-UA from UA-MTA
and MTA-MTA interactions. The specific proposal that you are making
merges them.
Acking delivery is a UA-MTA (final MTA) request and may possibly be
viewed as involuntary (a point for discussion) by the recipient.
Acking reading is very much by the receiving UA and is almost certainly
entirely at the discretion of the receiver.
So I suggest that we split the discussion into:
Sending UA and Receiving UA;
Sending UA and Mail Transport Service; and
Mail Relay to Mail Relay.
(There might be an argument for allowing MTS to Receiving UA, and one
could even argue that the RFC 822 Received headers fall into this
category, so this would be yet-a-fourth category of interaction.)
The assertion that Internet mail does not conform to the UA/MTA model
is, I believe, partly due to our tending to mix these mechanisms.
Standard packet net items:
1. Delivery failure detection;
2. Retransmission,
3. Congestion detection;
4. Flow control,
5. Data segmentation (fragmentation/reassembly),
6. Routing
7. Performance requirements.
The proposed ack mechanism touches item one. MIME message/partial touches
item 5, but we have no meaningful way to detect correct maximum message
sizes.
No routing feedback is given to senders. There was suggestion to have
a standard host for sending submissions to mailing lists, and one clever
person suggested having it send back a message indicating the correct
host, for future submissions. This is a classic "default router/ICMP
redirect" model and I rather liked it.
In other words, I'm suggesting that we attack these issues in a somewhat
systematic fashion, rather than entirely piecemeal. This need not delay
the specification of the pieces, but would, I think, be more inclined to
make them fit together better.
Question: Is it better to dribble such specs out or to issue an integrated
set, as long as it does not take too long to create a set?
Anyhow, grist for the mill...
Dave