lauren@vortex.UUCP (Lauren Weinstein) (10/10/85)
I should mention that one of my basic rules is that my domain handling
must BREAK NOTHING in the current environment, and must not REQUIRE
cooperating "smart" sites to work. In fact, I encourage people to
set up their tables to route mail DIRECTLY to a given site (regardless
of what domain it is in) when possible or practical. Any system
that "warps" traffic patterns for the sake of routing is going to have
trouble--instead, direct and semi-direct hops are to be encouraged
whenever possible. But as you'll note from my sample table in the
previous message, it's trivial to put in entries that go directly
to a given site or subdomain within a domain rather than relying
on any "domain server" or other very smart host except when
all else fails. It has the additional advantage of working regardless
of whether the sites around you are smart or dumb. The users and sites
that talk to your machine can address mail in User@Site.Domain form and
all the magic is taken care of automatically. They need not be
concerned about whether you will route to other smart sites or through
dumb sites. And after all, in the final analysis, it's user
convenience that's the point of all this in the first place.
--Lauren--
{ihnp4, decvax, seismo, clyde, trwrb}!vortex!laurenglc@akgua.UUCP (G.L. Cleveland [Lindsay]) (10/20/85)
As usual, Lauren points to a basic "truth" that sometimes gets overlooked: >From vortex!lauren >I should mention that one of my basic rules is that my domain handling >must BREAK NOTHING in the current environment, and must not REQUIRE >cooperating "smart" sites to work. I see the results of this quite often here at "akgua". Mail is sent thru here to "nextsite!joe@uucp" or "nextsite!joe.uucp", and "nextsite" has a vanilla-flavor "rmail" program to handle this. The result usually is that user "uucp" at my site gets the failure message from "nextsite" since it only understands the old "remote from" header lines and can't generate a more accurate return path. One tries to be helpful and the message is usually re-sent (manually) to the intended recipient using the old "bang" addressing. The usual culprit is the originating site with its "smart" address modifier. Bottom line...the lowest common denominator is the "bang" mail handlers. That method usually works on all of the uucp sites. (I'm not addressing cross-domain gateways here.) So as you debate methods and addressing standards, bear in mind "Lauren's Law" (slightly reworded): New mailers must BREAK NOTHING in the current environment, and must not REQUIRE other sites to change their software in order to work. Cheers, Lindsay Lindsay Cleveland (akgua!glc) (404) 447-3909 Cornet 583-3909 AT&T Technologies/Bell Laboratories ... Atlanta, Ga
lauren@vortex.UUCP (Lauren Weinstein) (10/21/85)
(First, an aside. Is it my imagination, or has SOMEONE out there suddenly dumped a lot of old netnews msgs back onto the net? I've recently gotten messages from people replying to articles of mine that were posted well over a month ago or more!) --- Well, I never intended my philosophy of "don't break existing mailers" to be codified as "Lauren's Law" (give me a break!) but if this concept were generally followed I suspect our network life would be somewhat easier. My concept is that there's nothing really wrong with sending fancy stuff down the line to another site (even, horrors, an address with an '@' in it) so long as you have sufficient information about the "state of the universe" to know how that address will be handled at various points along the path. For example, it is perfectly safe to pass '@' addresses directly to some ARPA gateway sites, in fact that's the only way to get some messages through the gateways. On the other hand, if you plan to pass such an address through old-style sites as well, or additional intermediate sites which also might take their own actions on the '@', it helps to understand the handling that each site will be giving that address, and to often NOT use the '@' (using instead the site.domain!user form) to pass through intermediate sites or to destination sites which understand that syntax. Yes, this philosophy requires information about how various sites will handle various addresses. And yes, keeping track of such information in our extremely hybridized environment takes some work. But it certainly seems better in the long run than just closing our eyes, gritting our teeth, and throwing addresses "blindly" down the line and hoping they won't screw up! --Lauren--
honey@down.FUN (Peter Honeyman) (10/22/85)
lauren's assertion to the contrary, mail routing does not \require/ information about the treatment of addresses at every site. furthermore, while such data might be useful, the classification problem would be intractable. for example, i use pathparse, so you might say that my treatment is the correct one (or you might not). but you can't categorize it as bangist or domainist or any other simple term. peter
lauren@vortex.UUCP (Lauren Weinstein) (10/23/85)
I've never said that mail handling REQUIRES knowledge of site "smartness." However, what I have said is that when you do have information about the smartness of sites it can allow reasonably intelligent mail handling without requiring "large" up-to-date databases or complex (relatively speaking) routing programs. In fact, with a bit of knowledge about the smartness of your most important neighbors the total amount of information you need may be very small overall. Knowing what you can send, in what form, to each of those sites can greatly simplify mail decisions. A site talking to 40 or 50 other sites can make reasonable decisions about how to handle a given @ or site.domain!user address with a very short table and only a few cycles. It doesn't take a great number of classifications (but there's more than just "domain" or "bang" to choose among). I currently have 3 primary classifications and about 4 options that can be attached to each of the three main classes. I'm not saying that having a full pathparse and mapping database and similar things doesn't allow equally good or maybe even sometimes "better" decisions, but, especially on smaller systems and systems that can't keep up with the mass of changing data for the universe at large, having information about "the smartness" of sites allows some useful tradeoffs. --Lauren--
peter@graffiti.UUCP (Peter da Silva) (10/24/85)
> New mailers must BREAK NOTHING in the current environment, and > must not REQUIRE other sites to change their software in order > to work. Which means that unless you know some site to be smart, don't give it any atsigns to deal with. INCLUDING in the From: address if you need to make one. ANY address you send out should be in either all-! or all-@ syntax.