Errors-to: owner-drums@CS.UTK.EDU
Message-Id: <199505312028.QAA12599@wilma.cs.utk.edu>
From: Keith Moore <moore@CS.UTK.EDU>
To: Robert Elz <kre@munnari.oz.au>
cc: Keith Moore <moore@CS.UTK.EDU>, Dave Barr <barr@math.psu.edu>,
        drums@CS.UTK.EDU
Subject: Re: getting started 
In-reply-to: Your message of "Thu, 01 Jun 1995 06:01:02 +1000."
             <824.801950462@munnari.OZ.AU> 
Date: Wed, 31 May 1995 16:27:57 -0400
Sender: moore@CS.UTK.EDU

> Along the same lines, and without being certain whether this
> is a cleanup, or new functionality in some sense or other, there
> badly needs, sometime, to be some kind of specification as to
> what all the Resent-* headers really mean, how and when they're
> used, etc.   That includes whether or not they should be used
> when sending replies, whether there is a Resent-Reply-To, ...

Agreed.

>     The 822 extension model tacitly assumes that since everything in the
>     header is for the benefit of the recipient's user agent, nothing you
>     put in a header should be able to affect transport or delivery.  This
>     needs to be stated explicitly.
> 
> I doubt you really meant to say exactly that.   If you did,
> this message wouldn't be able to get to you, as nothing would
> be allowed to look in the To: header to figure out where I
> intended to send it!

Of course if the sender is allowed to edit the bare headers, the
sender's UA has to be able to look at them to derive the initial
envelope.  But I don't think of this as "transport or delivery".

Yes, "sendmail -t" and other systems look at the header when
submitting a message.  In doing so they're just acting as part of the
sender's UA.  But it's absolutely incorrect for anybody except the
original sender to pipe a message to "sendmail -t".

Basically, 822 says you can add any field you want to the message
header, but you are taking the risk that it will conflict with some
other use of that field.  This works fine as long as the headers
aren't recognized by MTAs to affect transport or delivery, but things
start breaking once they start crossing that line.

It's also true that gateways may need to embed envelope information
somewhere for the purpose of tunneling, and sometimes the message
header is the best place to do this.  X400 gateways do this now.  It's
not a problem as long as the extra headers are only used for
tunneling, and don't directly affect Internet mail transport.  But
this is a slippery slope because casual users are likely to be
confused.

A clean separation between the user-agent protocol and the
transport-protocol is essential to the stability of the mail system.

Keith