Re: [idn] Using a new class for IDN

[John C Klensin <klensin@jck.com>]

--On Monday, 03 June, 2002 07:40 -0400 Edmon Chung
<edmon@neteka.com> wrote:

> Hi Dan,
> As an advocate for a clean approach to IDN I would like to
> offer my support. Though I think EDNS is a better solution
> because IDN is really just another domain name and should
> exist in both classes, I dont mind seeing a new class being
> created for IDNs because I believe that the world wants a good
> forward looking architecture for IDN and will move towards the
> new class as it becomes available.

Edmon,

A few observations:  The advantage of both EDNS and a new Class
approach is that they require explicit action on the part of the
part of the client to get "idn" treatment of a name.  Without
that action, and agreement from the server in some appropriate
form, nothing is going to happen, which means no heuristics
about how names are to be interpreted, no nonsense strings
showing up in front of users (at least with decent-quality
implementations), etc.  The disadvantage is what it always has
been and what caused the WG (I think) to ignore these solutions
from the beginning -- they are infrastructure changes, not quick
fixes, and the WG has been focused on quick fixes.

Between the two, I see a fairly similar tradeoff.  EDNS is
easier to think about in terms of incremental deployment, but
raises complex issues about, e.g., handling of secondaries.
"New class" gives us the opportunity --but also the
probably-very-painful obligation-- to rethink almost everything
and get it right this time.

But let me repeat, and expand slightly, the advice I previously
gave Dan in an off-list note.  My "new class" I-D was nothing
more than a sketch of how it might be done and what some of the
issues would be about doing it.  It does contain at least some
of the justification, which ought to be useful, but it is
(deliberately) notably short on fine details.  David Lawrence
has raised issues I didn't explicitly address as has, more
recently, Eric Hall.  What is going to be needed if anyone is to
take "new class" seriously as a proposal (and not as yet-another
attempted blocking action (of which I am not accusing anyone in
particular, only noting that there have been far too many
suspicions along those lines in the WG)) is a coherent internet
draft that really does the case analysis and considers, and
offers at least tentative solutions for, the many issues,
questions, and loose ends.

And, while I'll promise to read such a draft, I'm not going to
be much help.  IDNA and the various (incomplete) EDNS and new
Class proposals ultimately differ in how they represent data,
how the query stream is handled, and a number of other details.
While --being very conservative about risks to the DNS and the
Internet-- I still prefer the new Class one, I see far more of
the important problems lying in the areas of acceptable
character repertoires and matching rules for those characters
that are permitted, and I've concluded that _those_ issues and
people's expectations about them can't be dealt with in the DNS
at all.

So, if this really interests you, and Dan, and David, and anyone
else who has read this far, _please_ go form yourselves into a
design team and produce a comprehensive draft.  Get it posted.
And then have a discussion with Erik and Thomas as to whether to
try to get it reviewed in this WG or in DNSEXT, as well as about
its impact on the IDN/ IDNA proposals.  But, while I wouldn't
presume to speak for the ADs or the IDN WG, I can't imagine
anyone taking "new Class" (or anything else) seriously as an
alternative to the IDNA proposal at this late date unless there
is a clear and complete I-D on the table.  So, again, please
start writing I-D text, rather than notes to the WG mailing list.

And good luck.

     john