The conditions for changing the protocol (was: Mappings - some examples)

John C Klensin klensin at
Mon Nov 30 20:02:16 CET 2009

--On Monday, November 30, 2009 11:11 -0500 Andrew Sullivan
<ajs at> wrote:

> On Mon, Nov 30, 2009 at 08:05:50AM -0800, Erik van der Poel
> wrote:
>> By the way, one reason why HTML authors do not include Eszett
>> in their URLs is because MSIE6 does not support IDNA (and
>> there are still quite a lot of MSIE6 users).
> The above is a good argument why doing the right thing now is
> more important than getting backward compatibility perfect:
> we're still early enough in the IDNA support cycle that large
> numbers of clients don't do it, or do it imperfectly.  It will
> be way harder to fix things later.  Therefore, getting it
> right once even if it is more painful in the short term will
> pay off more handsomely in the longer term.
> (It also strikes me that this ground is well-tilled, but maybe
> not everyone who is paying attention now was paying attention
> the last time around.)

Yes.  And, of course, the "get it right now, and do it quickly"
position taken more or less formally in ICANN by the gTLD
operators and less formally by many of the ccTLD ones was driven
by exactly that reasoning.  Those positions, by themselves,
don't argue for or against classifying Eszett and Final Sigma as
PVALID or not, but they do stand in counterposition to the
"whatever was done in IDNA2003 must be preserved forever in the
name of compatibility".

The WG also considered, I believe, one other piece of data,
which was that, at least according to Google's statistics as
reported by Erik, the fraction of IRIs that contain characters
that map to others in IDNA2003 is on the decline.  Perhaps that
is because of concerns about IDNA2008, perhaps due to the same
realization about the confusion-minimizing advantages of
canonical names that the WG came to, perhaps for other
reasons... but, if one trusts in any of these statistics, one
showing a change in ratios should be, relatively, the most

If having those PVALID would have been The Right Thing to do in
2002/2003, then it is the right thing to do now unless there is
new evidence about the relevancy of the characters to IDNs.
Otherwise, even those of us who believe it would be better to
DISALLOW them now are faced with the risk of being forced (by
government action or the equivalent) to make them PVALID later.
Not only have we used words like "catastrophe" to describe the
circumstances that would justify that transition, but the
transition will clearly be harder a few years hence if IDNs are
more widely deployed than they are now.  And, of course, if we
don't expect IDNs to continue to deploy more widely over time
--significantly so-- than both this effort and the IDNA2003 one
will have turned out to be a waste of time.


More information about the Idna-update mailing list