Parsing the issues and finding a middle ground -- another attempt

[Andrew Sullivan]

On Wed, Mar 04, 2009 at 07:20:28AM -0800, Erik van der Poel wrote:

> One of the most important contexts for URLs/URIs/IRIs is HTML. In that
> context, I'd prefer to do no mapping, or, if really necessary, only
> /global/ mapping. We'd have interoperability problems if browsers
> started performing language-specific (local) mappings to domain names
> in URLs in HTML.
> 
> Local mapping might be useful in the URL bar, where browsers accept
> bare host names (without the leading "http://").

So the distinction I think you want to make is between (roughly) "user
types something in" (where local mapping might be allowed) and
"hostname in some other protocol-using context" (where local mapping
is not allowed).  Is that right?

If so, I'd like to note that the "user types something in" case is one
I'm really worried about.  We have enough complaints already about
confusability &c. without applications being explicitly permitted to
rewrite a substantial number of characters into absolutely any other
character or even combination of characters (for that's what "local
mapping" really means).

It's true that characters that are already covered by the protocol may
not be mapped.  But if there were no significant gaps in what the
protocol is covering, nobody would be arguing for mapping at all.
Since how those characters are to be mapped is to be a matter of local
policy, what we're really saying is that there is a class of
characters that will be typed by a naïve user into what it thinks is a
context for name resolution, and there is no way to tell just by
looking at the protocol what range of results the user might expect.

I did ask before that we not talk about "resolution" except in the
absolutely strict sense, and I'm delighted that we went in that
direction.  But I think we have an obligation to users to appreciate
that they will naturally model their expectations on what they know of
traditional DNS resolution, and I don't think we should do too much
violence to that model.  In my view, local mapping does a great deal
of violence to such a model, precisely because the results from
_exactly the same_ typing can be different in different applications,
on different machines, in different locale settings, or just from time
to time.

A

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