Re: [idn] WG last call summary

["D. J. Bernstein" <djb@cr.yp.to>]

Let me emphasize that the IDN WG has received public written objections
to IDNA from at least fifteen regular WG participants and _hundreds_ of
other people. Everyone agrees that the objections together accuse IDNA
of causing a tremendous amount of damage, including bounced email, web
link failures, widespread user confusion, and more.

The IDN WG chairs have nevertheless sent IDNA to the IESG. That's
outrageous. It is, within the IETF, the most blatant abuse of power I've
ever seen.

I'm willing to spend some time discussing the technical issues that led
to these objections; but I'd like readers to keep the context in mind.
These specifications should never have been sent to the IESG.

Now, back to the technical discussion.

Keith Moore writes:
> not everyone accepts your criteria for success.

``Internationalized domain names are a failure if non-ASCII glyphs don't
appear on the screen.'' What kind of idiot would disagree with that?
Would you like to explain, Keith, how IDNs can be a success without
non-ASCII glyphs appearing on the screen?

Paul Robinson asks whether I'm saying that _one_ bad display means that
IDNs are a _complete_ failure. Of course not. One bad display means one
failure. One billion bad displays mean one billion failures.

Obviously the IDNA authors understand the desire to put non-ASCII glyphs
on the screen. Why is this obvious? Because the IDNA documents say that
the ad-hoc IDNA character encoding ``SHOULD be hidden from users,'' with
various exceptions. The documents spend quite a bit of time explaining
how to convert strings out of the IDNA character encoding.

The problem is that this conversion creates interoperability failures.
Mail will bounce. Web pages will fail to load. The same failures would
not have occurred with ASCII domain names.

These failures are visible on the wire, and are caused by a proposed
IETF specification. The assertion that they're outside the IETF's scope
is patently absurd.

On Sun, 27 May 2001 21:30:52 +0000, IDNA coauthor Adam Costello claimed
on the IDN WG mailing list that nothing would actually break. On Thu, 19
Jul 2001 04:31:48 +0000, Costello admitted on the IDN WG mailing list
that his claim was wrong. Things will break.

To answer another question from Paul: The problem happens when the
converted data is fed to _any_ network program that isn't aware of IDNA.
It doesn't matter whether that program is an obsolete 7-bit application
or a modern 8-bit-clean application; what matters is that the program
isn't undoing the IDNA conversion.

When I give specific examples of these IDNA interoperability problems,
someone inevitably responds that we can fix those examples by turning
off the initial IDNA conversion. That is a silly response, for two
reasons:

   * Preventing _all_ the interoperability problems means turning off
     conversion for _every_ display subject to IDNA-unaware piping,
     IDNA-unaware copy-and-paste, etc.---and that's a massive failure to
     achieve the IDN goal.

   * The proposal on the table, IDNA, does not disable these
     conversions. It explicitly encourages these conversions.

Anyway, while some IDNA proponents obviously want to slither between
interoperability failures and display failures, Costello has already
admitted that IDNA will cause both types of failures.

---D. J. Bernstein, Associate Professor, Department of Mathematics,
Statistics, and Computer Science, University of Illinois at Chicago