Re: Injection-Date and reinjection

["Forrest J. Cavalier III" <mibsoft@xxxxxxxxxxxxxxx>]

Russ Allbery wrote:

> Since the normal case for disjoint Netnews networks is that they'll have
> disjoint sets of articles as well.  (Also, one doesn't need to limit it to
> one in each network necessarily.  I was trying to avoid making that
> limitation in my language.)

Simple changes.  Make it conditional... "To cause an article to appear on
disjoint networks a posting agent SHOULD..."

And "s/one/at least one/"

> > When it cannot be avoided, a posting agent which converts existing
> > articles back to proto-articles for reinjection to a disjoint network
> > MUST NOT reinject articles already accepted on that network.  It MUST
> > perform the same staleness tests of Injection-Date or Date header fields
> > as would be performed by a relaying agent.  (This helps prevent
> > reappearance of expired articles.)  It MUST NOT alter header fields
> > permitted in proto-articles, especially Message-ID.  (This helps prevent
> > loops.)  To make the article acceptable to an injecting agent, it SHOULD
> > rename other header fields to preserve information, but MAY remove them.
>
>
> Charles has a valid point that it's really beyond the ability of a posting
> agent to determine whether two Netnews networks are truly disjoint.

Yes, this point bothered me as I was going to bed.  Relays ought to check Path
and would rewording indicate that?  Would this be adequate.

> Beyond that, the bit about the Message-ID is a good addition that I want
> to keep, regardless of how else we word this.  The gateway stuff already
> says this sort of, but we should emphasize it.
>
> We need to say *what* other header fields it will rename.  I think we also
> have to allow for drop as well as rename; for example, if I'm posting to a
> disconnected INN server and then gatewaying those articles to Usenet, my
> local trace information is entirely irrelevant and there's no reason to
> retain it.

Isn't that permitted by the last SHOULD?  Why is the trace information
entirely irrelevant?  It isn't any more irrelevant than opaque trace
information from all the other servers.

> If this is the complete section, we're not saying that the gatewaying
> rules apply anywhere, which I think we still need to do since there may be
> bidirectional gatewaying at work (among other things).

Unnecessary complexity.  Why make people at private leaf nodes consider
the requirements of bidirectional gatewaying, discarding half of those
requirements?

Can't bidirectional gatewaying be discussed under gatewaying, or let
implementors assume it is just two unidirectional reinjections?  The bit about
using the Path header may be important for this case.  What concerns
do you have?