From: Russ Allbery (rra@stanford.edu)
Date: Sat Apr 29 2000 - 23:53:27 CDT
Charles Lindsey <chl@clw.cs.man.ac.uk> writes:
> If the article is rejected, then it fails for all the newsgroups
> for which it was intended (in particular the moderator MUST NOT
> resubmit the article, with a reduced Newsgroups header, to any
> remaining groups). If the article is accepted, the moderator
> proceeds with the following steps.
SHOULD NOT. Not trimming the newsgroups header is a matter of policy; we
can recommend a policy, and I think the interoperability concerns warrant
that, but it is possible to do this in a way that does not break news, and
there are existing moderated newsgroups for which the moderator regularly
modifies the Newsgroups header.
If we make the switch to OUGHT, I think this should be an OUGHT NOT rather
than a SHOULD NOT.
> 2. The Date header MUST be retained, except that if it is stale (5.1)
> the article SHOULD be rejected.
The Date header SHOULD be retained, except that if it is stale the
moderator may replace it with the current date if appropriate (such as
if the delay in injection were due to moderation delays). If the
moderator replaces the Date header, the moderator takes responsibility
for ensuring that loops are not created, as overriding the Date header
may break the ability of the news system itself to do this.
Replacing the Date header to prevent premature expiration or injection
difficulties when the delay is due to moderation delays is a not-uncommon
practice and shouldn't be in violation of our standard.
> Incoming gateways MUST NOT inject control messages other than
> cancels. Encapsulation SHOULD be used instead of gatewaying, when
> direct posting is not possible or desirable.
> NOTE: It is not unheard of for mail-to-news gateways to be used
> to post control messages, but encapsulation should be used for
> these cases instead. Gateways by their very nature are
> particularly prone to loops. Spews of normal articles are bad
> enough; spews of control messages with special significance to
> the news system, possibly resulting in high processing load or
> even e-mail sent for every message received, are catastrophic.
> It is far preferable to construct a system specifically for
> posting control messages that can do appropriate consistency
> checks and authentication of the originator of the message.
> [Is that really so bad? I do it regularly.]
In addition to the worry about spews, there's also the various problems
caused by the transformations necessary for a gateway and the signatures
on control messages. It's a whole pile of potential problems. I'd rather
stick with the MUST NOT and require that if it's absolutely necessary for
control message injection to happen via e-mail, that it be by special
arrangement and not a general gateway.
> Exceptionally, if there are multiple incoming gateways for a
> particular set of messages, each incoming gateway SHOULD generate a
> message identifier unique to that gateway. Each incoming gateway
> nonetheless MUST ensure that it does not gate the same message twice.
> [Should that SHOULD be a MAY?]
There are interoperability problems with not using a separate message ID
space for each separate newsgroup that gateways a given mailing list,
namely that if one does not do this, each newsgroup will only get a
portion of the traffic. They're certainly less than the problems of
loops, though.
-- Russ Allbery (rra@stanford.edu) <http://www.eyrie.org/~eagle/>