From: Bryan Reece <reece@taz.nceye.net>
Date: Tue, 18 Feb 1997 06:18:02 -0800 (PST)
To: "E. Allen Smith" <EALLENSMITH@ocelot.Rutgers.EDU>
Subject: Re: Anyone have the complete info on CP list alternatives?
In-Reply-To: <01IFJH1ONXA88Y4YVV@mbcl.rutgers.edu>
Message-ID: <19970218141819.28530.qmail@taz.nceye.net>
MIME-Version: 1.0
Content-Type: text/plain


E. Allen Smith writes:
 > From:	IN%"reece@taz.nceye.net"  "Bryan Reece" 17-FEB-1997 23:23:10.06
 > 
 > >>   use. However, I do have one problem with this recipe: lack of
 > >>   loop prevention if a news2mail gateway is going in the other direction.
 > 
 > >All the mailing-list hosts are supposed to check for duplicates before
 > >forwarding anything, so this shouldn't do anything worse than have a
 > >few extra copies of each post get discarded, loading the network and
 > >the hosts a bit more.  But posting the list to a widely-distributed
 > 
 > 	Umm... IIRC, the only person who's announced how he's filtering
 > for duplicates is Igor, and he's just checking for messageIDs. Won't
 > those get chunked on going through a mail2news gateway, and thus
 > possibly come back through a news2mail gateway?

My script preserves the Message-ID, assuming that the incoming message
has one.  Other gateways I've seen also appear to preserve Message-ID
lines (or at least messages posted through them appear to have
mail-like ID lines).  

 > The filtering
 > mechanism could be improved via MD5 digests, etcetera (someone's
 > already given a recipe for such; thank you), but even so various
 > mungings could still set up a mailing loop. (Yes, I'm paranoid
 > about those; I may have been reading list-managers for too long and
 > seen too many stories on there.) The sensible place to keep
 > track of possible duplicates is at the gateway; it can
 > certainly filter based on that it (or, for that matter, another
 > known bidirectional gateway) sent a message out.

Not immediately sending a message back to the host that it came from
does look like a good idea (it came from foo.com, so foo.com obviously
has it).  Filtering out other gateways' messages will reduce load at
the possible expense of delay (assuming a smallish network of news
servers that feed each other the list; obviously the message will
flood the majordomo net before it floods the real usenet unless
something is seriously broken).

 > 
 > >newsgroup seems wrong for noise (both to the list itself and to
 > >posters) and delay reasons.
 > 
 > 	Delay reasons being that mail will go to Usenet, go
 > all over the place there, and then get replied to later than if
 > it were just on the mailing list(s)? A potential problem, yes.

I didn't exactly say, but I was comparing a small network of news
servers feeding each other cypherpunks.list to, say,
alt.cypherpunks.list. 

 > But (especially given the options I mentioned below), this
 > would appear to be a matter for individual users to decide
 > by which list they go with and any added control headers.

I'm trying to come up with a way to provide NNTP access (and an
experimental alternative to the net-of-majordomos) that won't offend
anybody too much.  I can filter out messages that the authors don't
want sent out; I'd prefer to deal with what would make the authors
want the messages filtered out instead.

 > Maximum individual sovreignty (sp?).
 > 
 > >>   In regards to your later query about whether people want their postings
 > >>   going to Usenet, might I suggest that this be individual to the given
 > >>   distributed mailing list? In other words, since the recipe is logically
 > >>   going to have to not forward to a gateway messages from other mailing
 > >>   lists (since it's not a good idea to have multiple copies of the same
 > >>   message arriving at the gateway if it can be helped; better to filter
 > >>   it out beforehand),
 > 
 > >It doesn't seem an especially bad idea, since it 
 > 
 > 	Thank you. Were you cut off?

Yes.  Should have read:

It doesn't seem an especially bad idea to send a gateway multiple
copies of the same message, since theextra messages won't increase the
load much but will increase the reliability and decrease the
propagation time compared to a network where there are no duplicates.
The only case where filtering duplicates is certain not to hurt delay
or reliability is when the gateway in question is known to have a copy
of the message already.  Usenet uses the Path header for this; there
doesn't seem to be such a thing in the majordomo net.

 > >>   some of the mailing lists can forward and the
 > >>   others cannot. (One could even determine this behavior on application
 > >>   of the proper X-header, although I never trust various mailing systems
 > >>   to forward such intact.) The same could also be done with gatewaying
 > >>   _from_ Usenet - if the news2mail gateway feeds to whatever individual
 > >>   lists sign up to it and they _don't_ forward such messages to others,
 > >>   people can decide whether or not to receive Usenet postings on
 > >>   alt.cypherpunks.* by which list they subscribe to.
 > 
 > >The code currently running is a perl script that turns typical email
 > >messages into something INN is happy with.  It is probably possible to
 > >get your message rejected by INN if you put obsolete or otherwise
 > >unusual and illegal headers in.  This may be a bug.
 > 
 > 	That would probably depend on: A. if any other circumstances
 > other than deliberately sabotaging your message's translation would
 > also disrupt its chances of getting through seriously, particularly
 > if they weren't something you'd spot; and 

The main way to get a message rejected seems to be using obsolete
news headers on it.  If you don't want your message posted, add a
Posting-Version line to the header.

 > B. if sufficiently bad
 > messages would disrupt INN's operations too much.

No more so than a sufficiently bad news posting that came in over NNTP
(since the mails do turn into news postings over NNTP).