Path: utzoo!mnetor!tmsoft!dptcdc!jarvis.csri.toronto.edu!neat.ai.toronto.edu!rayan From: rayan@ai.toronto.edu (Rayan Zachariassen) Newsgroups: comp.mail.sendmail Subject: %-@ formalization (was Re: sendmail parsing questions - my comments) Message-ID: <89May13.164744edt.11600@neat.ai.toronto.edu> Date: 13 May 89 20:47:36 GMT References: <1635@ur-cc.UUCP> <53257@uunet.UU.NET> <89May11.183003edt.11589@neat.ai.toronto.edu> Organization: Department of Computer Science, University of Toronto Lines: 100 In article <32716@sgi.SGI.COM> vjs@rhyolite.SGI.COM (Vernon Schryver) writes: # What should be done done with a!b!c!d!e ? More than one % ? Keep some of # the !'s ? Bounce it if the next hop, 'a', cannot be reached by UUCP, even # though we are of course assuming it is a valid, registered, approved domain # (or has such an synonym according to the maps)? I've seen more than a # couple of such relatively benign things pass both ways through SGI's # gateway. # # I'm just asking for the revealed wisdom of the coming RFC and what the # big gateways will do, and not intending to attack anybody or anything. I'm not in a position to reveal wisdom about the HR RFC, you should ask Jon Postel about wisdom questions... With that disclaimer, here is what I've done in the past and how I view the present and future in this matter: About 3 years ago I wanted to start mailing to and through (we were on CSNET) the Internet, and wanted to put knowledge of Internet-UUCP/BITNET/... gateways into the pathalias data we use for routing. After some experiments I found that 90% of sites at the time could handle @a,@b,...:u@h format without being overly critical of what was at the RHS of @'s, and about 70% knew about %-@ but required a particular syntax for the name between %-@, or would prefer ! to % or some other idiosynchratic behaviour. So because I didn't want to keep track of which hosts wanted what format, and as the lesser of evils, I settled on using '822 source routes for everything; i.e. I translated UUCP source routes into the equivalent '822 syntax, consciously violating the semantics ("must be Internet domain name"). There were few sites at the time that would not accept this, I vaguely recall decwrl and ucbvax being the major culprits (from this point of view). Since then the situation seems to have changed a lot. People heard of '976 and, along with new Sendmail releases, the mailers became more restrictive in what they would accept and how. Now some of the major gateways seem to follow the letter of '822 wrt RHS of @, and usually implement '976 so that they understand uucp-neighbour!u@myself.dom. However each site is still idiosynchratic in what else it accepts; for example inet-host!u@myself.dom or u%uucp-neighbour@myself.dom or u%uucp-neighbour.uucp@myself.dom or uucp-neighbour.uucp!u@myself.dom may all fail. In theory it is possible to get any syntax through an address-rewriting mailer by quoting the next-addr. I.e. even though a mailer would normally change a!u@h.dom to @h.dom:u@a.uucp (for example), one can always say "a!u"@h.dom to force the h.dom mailer to route a!u instead of u@a.uucp. In practise there are very few mailers that deal properly with quoted local-parts, so this doesn't work too well. The situation now is chaotic. I argued with the HR RFC group to change the semantics of @ RHSs, so people wouldn't be skittish about translating non-Internet source routes into '822 format. Instead they (and I now believe this is probably the right approach at present) are thinking of formalising the %-@ hack. Within the Internet then, the RHS of @ will always show an Internet domain name. Each mailer on the Internet will be responsible to maintain this: turning u%a@b.dom into u@a.dom. So far so good, we will have a mechanism using known syntax to express non-Internet source routes. What I think is missing is for all the Internet-net gateways to agree on how this syntax will be used... now we get to Vernon's first few questions. All the answers I'm giving below is what I would like to see happen -- personal opinion follows: # What should be done done with a!b!c!d!e ? More than one %? Yes: e%d%c%d@a.dom # Keep some of the !'s ? No. This is the situation we are in now. Personally I don't want to see an ! anywhere in Internet addresses unless the powers that be choose to go with UUCP-style source routes and deprecate @: and %-@ completely (unlikely). In other words, the gateways should do a complete transformation between ! and %-@ format. Note that even if you do keep some of the !'s it is still 976 conformant, but you can't guarantee that the mailer you'll be sending it to ever heard of 976 (or perhaps it implements adaptive !-%-@ prioritization and gets it wrong occasionally). # Bounce it if the next hop, 'a', cannot be reached by UUCP, even # though we are of course assuming it is a valid, registered, approved domain # (or has such an synonym according to the maps)? No. Even though the ! syntax started out for uucp-only links, nowadays it is very counterproductive to insist on a specific syntax for specific links. We should pay attention to the semantics: both a!... and ...@a mean "get it to 'a' somehow, I don't care if its by courier pigeon or on a moonbounce laser". This still doesn't address what to do with .uucp or .bitnet or other pseudo-domains. As for what I intend to do... wait for the HR RFC to come out and see how the mailers on other sites change to support %-@ better. If appropriate I'll then spit out %-@ to the Internet. For all you gateway admins: is there any interest in collaborating on a follow-on RFC dealing with this view of address translation at internet gateways? The objective might be to stamp out ! and :: on the Internet and to have agreed-upon semantics for the addresses you put in %-@ form. rayan