Date: Mon, 22 Sep 1997 21:50:33 -0400 (Eastern Daylight Time)
From: John C Klensin <klensin@mci.net>
Subject: Re: TURN and disconnected SMTP with dynamic IP addresses
In-reply-to: <199709230009.UAA28233@snark.thyrsus.com>
To: "Eric S. Raymond" <esr@snark.thyrsus.com>
Cc: jeffstep@microsoft.com, drums@cs.utk.edu
Reply-to: John C Klensin <klensin@mci.net>
Message-id: <SIMEON.9709222133.C@dataarch1.mci.net>
MIME-version: 1.0
Content-type: TEXT/PLAIN; CHARSET=US-ASCII
Priority: NORMAL


On Mon, 22 Sep 1997 20:09:08 -0400 (EDT) "Eric S. Raymond" 
<esr@snark.thyrsus.com> wrote:

> ...
> I maintain a POP/IMAP client (indeed, as of right now I would bet it
> is the single most popular POP/IMAP client in use on Unix
> machines). POP and IMAP are attacks on the "disconnected-host
> problem".  Your `separate, carefully-designed and authenticated "send
> me all the mail for FOO"' format *is*, in fact, IMAP4.  Or should be.

If I were a little smarter, I probably wouldn't touch this 
one.  But I'm obviously not, so:

No, it isn't and it shouldn't be.  IMAP4, and POP3, are 
designed around a split-UA model of the world in which the 
transport system delivers to a maildrop and those protocols 
are used to retrieve things, more or less selectively, from 
that maildrop (POP3 is obviously a lot less selective than 
IMAP4).  It is inappropriate to think about such things as 
mail gateways or anything else that might involve rerouting 
or translation functions.   What you are trying to do is to 
turn them into "transport pull" protocols, which are a 
different sort of critter.  So, indeed, they do rather 
poorly things that they were never intended to do.  This 
should not be a surprise.

> Bzzt!  Wrong answer -- or, rather while narrowly correct, it fails to
> address the larger design problem here, one my practical experience as
> the fetchmail maintainer continually grinds my nose in.  (We appear to
> have a historical problem here in that the POP/IMAP community hasn't
> generally talked with the SMTP community much, or vice-versa. Well --
> consider me Boy Crossover...)

Quite the contrary: the active designers of POP and IMAP 
have been very much involved with SMTP and, at least in the 
last several years, vice versa.  Again, you are trying to 
solve a different problem -- one best addressed by SMTP 
extensions (maybe) or some sort of mailbag protocol, or 
some transport protocol that more closely resembles UUCP or 
BITNET BSMTP (which are both closer to mailbag protocols 
than SMTP is).

>...
> "This is a bad idea" is the wrong answer.  *Something* logically and
> functionally equivalent to a full-fledged mailbag format has to happen
> in order for the disconnected-host problem to get really solved.  The
> security and disclosure problems with such an approach are there *but
> we'd have to solve them anyway*.

We probably agree on this.  Our disagreements, if any, are 

  -- whether it is desirable to attach such a protocol to 
    POP or IMAP
  -- whether it is desirable to attach such a protocol to
    SMTP
  -- whether hacking at TURN is an appropriate way to fake
     such a protocol

To semi-answer your last question, I'd love to see serious 
work on a mailbag protocol, and would be willing to spend 
some time on it, but believe that it is important to get 
DRUMS finished and have a clear foundation first.

--john