[NNTP] Re: New NNTP drafts approaching IETF Last Call

[Russ Allbery]

Clive D W Feather <clive at demon.net> writes:
> Russ Allbery said:

>> I was hoping to avoid NNTP having to be the standard that had to
>> specify the stringprep for newsgroup names, just because that's going
>> to be a zoo to nail down and to some degree it's premature because no
>> one's really done the hard work on i18n newsgroup names yet.

> *I* did a load of hard work in i18n newsgroup names a few years ago.

Yes, that's very true, and I'm sorry, I hadn't meant to imply that you
didn't.  What I was trying to drive at was that we didn't have an RFC or
even an I-D that looked likely to become an RFC, we didn't have much in
the way of implementation experience, etc.

You did indeed lay the groundwork, and I wish that would have been more
successful.

> I don't know what has survived, because I left Usefor soon after, but
> one thing that was clear was that people aren't willing to do
> normalisation every time an article or newsgroup name wanders past.

It's not clear to me that this would be something about which we have a
choice.  In following the IETF WG chairs mailing list and some of the
other IETF policy mailing lists, i18n and character set issues appear to
be matters that have increasingly been nailed down by i18n-specific
working groups and applied IETF-wide to all protocols.  Individual
protocol working groups are getting less and less leeway in this area and
are being strongly encouraged to simply reuse the efforts of the i18n
experts, due to the wealth of complex and difficult issues involved in
handling this properly and securely.  (I am generally in favor of this
trend.)

Obviously there are optimizations that one can make internal to news
clients.  If the news client already knows that the newsgroup name is in
canonical form, it doesn't have to recanonicalize it every time it's used.

> The best model was for there to be a canonical form for newsgroup names,
> for the newgroup message to enforce that form, and for everyone else to
> assume it had already been done. So LIST ACTIVE, for example, would only
> return canonical names.

This is certainly the model that makes the most sense to me, but I'm also
painfully aware of my lack of expertise in this area and would have to
defer to you, IETF i18n experts, or anyone else who has done significant
work in this area and hammered out the various issues involved.

-- 
Russ Allbery (rra at stanford.edu)             <http://www.eyrie.org/~eagle/>