Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!watmath!clyde!rutgers!im4u!ut-sally!ut-ngp!infotel!pollux!ndmce!tp
From: tp@ndmce.UUCP
Newsgroups: comp.mail.uucp
Subject: Re: From: lines ...
Message-ID: <1166@ndmce.uucp>
Date: Wed, 28-Jan-87 10:52:51 EST
Article-I.D.: ndmce.1166
Posted: Wed Jan 28 10:52:51 1987
Date-Received: Fri, 30-Jan-87 05:06:51 EST
References: <1139@ndmce.uucp> <16950@ucbvax.BERKELEY.EDU> <1159@ndmce.uucp> <2862@soma.bcm.tmc.edu>
Reply-To: tp@ndmce.UUCP (Terry Poot)
Organization: Nathan D. Maier Consulting Engineers, Dallas
Lines: 103

In article <2862@soma.bcm.tmc.edu> sob@cortex.UUCP (Stan Barber) writes:
>It should be noted that smail does enforce conformance to RFC 822 in that 
>it does not supply missing but necessary headers in mail that is missing them.
>smail is mainly an router as is uumail. sendmail (and mmdf, I think) enforce 
>the RFC by supplying those headers if they are missing. Therefore, it is 
>overstating the capabilities of smail to say that it conforms the RFC 822
>(or RFC 920 or RFC 976).

Good point (I assume you meant does not above). So, does anyone who does
not run sendmail have either a user-agent, or something between the
user-agent and router, that will do RFC822-compliant header processing?
For instance, I know my mail leaves here without a message-id, and I
believe this is illegal (can't find my copy of RFC822 right offhand). It
would be nice to have a user agent that complemented smail's capabilities.
I use mailx sometimes, and mail sometimes, because mailx thinks that any
domained address needs to have .UUCP appended. My system routes messages
just fine. Now I need a decent user-agent. *Sigh*

As far as sendmail doing the headers, it seems that there are mixed
opinions as to whether it can be made to do it right. Jordan Hayes says it
can, but others (forgot who, sorry) say it can't without hacking the
source. I'd heard that before, and that was what I was referring to in the
message that started that argument. Several people have sent me
configuration files that they claim do it right, but it will be a while
before I learn enough about sendmail to be able to tell.

What I'd like to get my hands on is several programs that each implement a
'layer' of what needs to be done. Sendmail already does many (or most) of
these things, but the jury is still out on whether it can be made to do it
completely right, which indicates to me that even if it can, it isn't easy. 
I'd enjoy seeing a discussion of the different facilities needed, how they
should be divided among the different 'protocol layers', etc. (The ISO
model might not even be a bad place to start, although I guess to match
reality, we'd have to drop the layers that make sure the message actually
gets where it is supposed to :-)

For instance, you need :

1) A user agent (of course). Dave Taylor's ELM seems quite good, but it 
does its own routing (i.e. doesn't restrict itself to being a user agent,
so you can't plug in things like smail or uumail (v4). One complaint I've
heard about ELM is that it is so large it takes a long time to start. So
there is obviously a lot of personal preference in what people would like
to see in their user agent (small and quick vs large and powerful being
just the most obvious).

2) Routing. Figuring out how to get it there. smail seems to cover this
well. I haven't seen uumail v4, but I'm sure it does quite well at this
too.

3) Transport. UUCP is less than ideal. Sendmail augments this greatly by
setting up a queue and being intelligent about managing it. If it can't
deliver it, it gets returned to the sender. On my system, it gets bounced
back only to the previous machine, which is not good.

The problem, of course is that there are pieces missing, and no standards
to cover them. Also, there are functions that are hard to relegate to just
one of these areas. Aliasing can meaningfully be done at least at the user
agent and transport levels. Also, in the uucp world especially, there is
inherent aliasing going on during routing. In the domain-based view of the
world, the uucp name of a machine would have to be considered an alias.
(Just think, if EVERYONE used pathalias, all we would have to put in the
maps would be domain names, and only your neighbors would have to know
about machine names. Duplicate machine names wouldn't be a problem. Of
course we'd have to worry about the length of the rmail command line with
all those 20-30 character site names!)

The problem that seems to be of largest concern right now is getting the
routing and transport layers to work together properly, and deciding just
what each needs to do.  I think that there is a lot that I don't understand
in this area, as I can't see why it is as difficult as it seems to be.  For
example, I don't see why you can't ask the internet name server, and if it
isn't on the internet, send it via uucp.  This however appears to be a
major problem.  I also don't see why any site that is connected to the
arpanet and runs pathalias can't forward any arpa mail out to uucp.  If it
isn't on arpanet, hand it to smail.  Note, I am not saying it is this easy,
if it were it would have been done.  But perhaps a wider discussion of the
problems might lead to a solution. 

Also, the problem is bigger than that, with LANs, other networks, etc.
They tend not to have name servers, so how do you tell where they are. And
of course nameservers in that sense are impossible in uucp, unless you are
willing to wait several days for the response to the query.

Most of the discussion thus far has concentrated on routing, and I think
perhaps a lot of the problems thus far arise from that fact that not
enough attention has been paid to transport, and it's inter-relationship
with routing. Note that the only person who claims to have solved the
transport problem to his satisfaction has done so by adding information to
the routing data (pathalias entries) to force the transport information to
be specified as an explicit part of the route. Is this the way to go? If
so, what would we need to support it net-wide?

Sorry to have all the questions and none of the answers, but I don't have
sufficient understanding of the transport problems to propose any
solutions (being a uucp only site). The alias problems are pretty easy (I
just have a shell script pretending to be rmail that handles that well).

-- 
Terry Poot, Nathan D. Maier Consulting Engineers, (214)739-4741
8800 N. Central Expressway, Suite 300, Dallas, Tx 75231, USA
UUCP: { seismo | cbosgd | ihnp4 | sun!convex | allegra!convex }!ndmce!tp
ARPA: ndmce!tp@seismo.css.gov   CSNET: ndmce!tp@smu