Comments on draft-ietf-idnabis-defs-10

[Andrew Sullivan]

On Mon, Aug 31, 2009 at 10:58:53AM -0700, Paul Hoffman wrote:

> Andrew, a developer who actually read the Punycode RFC, missed the subtlety.
> 
> Wil alerted the list months ago, and all of us still missed this subtlety.
> 
> It would be an act of absurd optimism to think that developers will
> get this without us being very explicit in both documents about what
> we mean.

In my case, it's actually worse: I waded into the thread Wil started
months ago precisely because I was worried about what he was raising &
wanted to avoid having it conflated with another, unrelated problem.
Alas, I think I contributed to the confusion instead of keeping the
discussion focussed.  But I still missed this in my review, much to my
embarrassment.  

Therefore, I do think we need to make this issue completely plain, but
I am also still not sure whether there is anything to change here.  If
I understand the situation correctly, there is no such thing as an
A-label that includes an upper-case ASCII character.  This is because
A-labels are defined in terms of being convertible to and from
U-labels, and there's no such thing as a valid U-label that includes
those upper case characters.

Now, there's an interesting upshot of all this, and that is that every
A-label also compares as equivalent to some number of R-LDH labels;
the resulting set of labels is every permutation of the A-label in
question with the ASCII-range characters uppercased.  None of those
labels are A-labels, because they don't obey the restriction that they
can be produced by conversion from a U-label (because no U-label may
contain such characters).  So far, however, this sounds to me like a
note that ought to be highlighted, but not a formal part of the
definitions or protocol.  I'm still open to persuasion otherwise.

A

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