Stop me if I've misunderstood...

[Kenneth Whistler]

> At 7:02 PM +0000 7/10/09, Shawn Steele wrote:
>> I agree with Mark that browsers MUST map, otherwise you have chaos, 
>> however I'm not sure I want to say "X, Y & Z" must map and "A, B & C" 
>> must not map.  I think saying that some applications SHOULD map 
>> is sufficient.

and Paul Hoffman responded:
 
> Stop right there. Read the definition of "SHOULD" in RFC 2119.

O.k.:

3. SHOULD   This word, or the adjective "RECOMMENDED", mean that there
   may exist valid reasons in particular circumstances to ignore a
   particular item, but the full implications must be understood and
   carefully weighed before choosing a different course.

> Come up with a proposed sentence (not idea) that expresses what
> you want that meets the RFC 2119 rules.

There may exist valid reasons in particular circumstances for
some applications to not casefold domain names, but the full
implications of not casefolding must be understood (i.e., that
there may be other strings, with different casing of letters,
which if casefolded, would resolve to the same DNS record)
and carefully weighed before choosing not to casefold.

The browser implementers seem to be saying (and I agree) that
they would always casefold, but if we define casefolding as
a SHOULD for the protocol, as I read RFC 2119, that would
not then be a MUST (i.e. an absolute requirement of the
specification).


> We can't rely on RFC 2119 and then blow off its semantics.

Fine, but...


> Then you cannot use "SHOULD" in the RFC 2119 context. So, some of 
> "we" very much want to go there, or not use RFC 2119 language. 
> MUST is easy, as is MAY.

I don't see "MAY" as at all easy in the context of discussing
what to do for IDNA 2008, because:

5. MAY   This word, or the adjective "OPTIONAL", mean that an item is
   truly optional. ... An implementation which does not include a
   particular option MUST be prepared to interoperate with another
   implementation which does include the option, though perhaps
   with reduced functionality....
   
And what Mark and Shawn and Gervase seem to be saying, at least
as I understand it, is that if you do not require casefolding
(and the other kinds of mapping needed for at least minimal
backwards compatibility with IDNA 2003), then you simply do
not have interoperability. You have different browsers resolving
to different locations, based on optional choices and levels
of IDNA library support, among other things. And if you
can't guarantee interoperability of implementations (though
perhaps with reduced functionality), then you cannot make
such an element an OPTIONAL part of the specification.

I just don't see any way that we can have casefolding (at
least) be a "MAY", and end up with a coherent, usable
protocol that will actually be implemented and used.

--Ken