Implementation questions (digressing from...)

[Erik van der Poel]

On Tue, Dec 23, 2008 at 12:00 PM, Shawn Steele
<Shawn.Steele at microsoft.com> wrote:
> Erik said:
>> Absolutely. However, the A-labels corresponding to U-labels that
>> contain eszett, final sigma, ZWJ and ZWNJ do not currently work in
>> IE7. We just have to pray that Microsoft will fix that, and that
>> users' copies of IE7 will be updated. If so, we don't need another
>> prefix.
>
> I'm not sure that prayer will be as effective as sending someone
> at Microsoft an email (like me) :)  I would hope that I don't seem
> unapproachable.

I apologize for wording it that way. You are a very approachable
person, and I am grateful for your response.

> From a browser perspective, it seems that if I encountered an
> eszett I'd have to use the 2008 rules, and if those don't succeed,
> fall back to the 2003 rules.

There is another alternative, and I wonder what you think about it. If
a browser will continue to "pre-process" even after adopting IDNA2008,
then eszett is always mapped to ss and the only way for an HTML author
to actually include an eszett is to insert it into a U-label and then
compute the IDNA2008 A-label from it, inserting that A-label into the
HTML. One rather important advantage of this approach is that neither
browsers nor crawlers need to ever do two DNS lookups for a single
domain name. Of course, one of the disadvantages is that one cannot
have a raw eszett in the HTML and expect the browser to refrain from
mapping to ss. At the moment, my opinion is that the stated advantage
outweighs the stated disadvantage. (But there may be other advantages
and disadvantages.)

> Just to randomize the conversation:  I can see where ß and ss
> can differ linguistically, but in practice I can't see how they can
> resolve to different domains.  <CrazyIdea>So it seems like I (as
> a domain owner) need the distinction primarily for display, not
> for resolution.  In other words: how about allowing PTR records
> or maybe a special CNAME or something that resolves a name
> to its preferred display form, undoing any mappings that were
> encoded?</CrazyIdea>

I kinda like where this might be headed. I would hope that the display
preference could be returned together with the IP address, and that it
would use a mechanism that is guaranteed to be passed all the way back
to the client. Would either of your suggestions (PTR, CNAME) provide
this?

Erik