Stop me if I've misunderstood...

[Andrew Sullivan]

On Wed, Jul 08, 2009 at 10:22:55PM +0000, Shawn Steele wrote:
> 
> There is one, singular, context, and that is as Domain Names that
> get you to a server.  That is consistent.  Any other context isn't
> IDN.

I believe the above premise was exactly the sort of thinking the panel
who presented the initial thinking that led to idnabis were rejecting
when they presented this topic in (help me out, panel -- Prague?).  To
put it melodramatically, the original idea was for a uniform
experience, uniformly bad for everyone all the time.  The claim in
$meeting was (as I understood the presentation) that such a principle
was bad.  The new principle was closer to "localization", and the goal
was to make things work more or less reliably more or less all the
time for the more or less natural cases, and accept that the
interstices were going to be ugly.  My impression at the time was that
the argument depended on the premise, "The interstices are ugly
anyway, so we're losing nothing."

Even though I don't remember the city where we had this conversation,
I remember going to the mic (in front of the entire IETF crowd, and
with some trepidation) and asking whether this approach didn't
inevitably result in a kind of nasty linguistic fragmentation of the
namespace.  The answer that I remember was, "Yep, and if you think you
don't already have that then $sales_offer_too_good_to_believe."  Those
who presented on that panel, and who have subsequently done the heavy
lifting, can contradict me, since I freely admit my memory is faulty.

But my overall impression of the direction that we have been going is
that everyone almost-cheerfully admits that the IDNA2008 answer does
not solve all problems, but that it solves some problems that were
previously dismissed as non-problems (or, to put it more neutrally,
were problems that were just not solved, and were ignored, or rejected
as too hard, or not problems yet) when IDNA2003 came out.

What the WG is charged to find consensus on, then, from my (personal,
obviously) point of view, is just where the "sweet spot" is for the
compromises.  I don't know the final answer, but I am not convinced
that "no local context" is one of the options on the table; if it
were, the WG wouldn't have been needed, because exactly the
IDNA2003+latest-unicode approach that Paul Hoffman proposed would have
been the only option open.  

By the way, none of the above is an attempt to dismiss your worry.  I
agree with you.  Localization in a global context without other clues
is plenty hard, and fraught with lots of traps.  The response to that
so far is, "Zone managers[1] need to have coherent policies, and the
default is always, 'No.'"  For this reason, I think the basic approach
of the "mapping" document is right, even though I understand the
interoperability worries you (and, I think, Mark) are raising.  Zone
managers will just have to be more active.

A

[1] Note that there are DNS-ish arguments against that name, but I
don't want to get into those weeds right now.  We can refine
terminology at the last step.

-- 
Andrew Sullivan
ajs at shinkuro.com
Shinkuro, Inc.