Re: [idn] Reality Check

["Adam M. Costello" <amc@cs.berkeley.edu>]

[A response to Mats Dufberg appears at the bottom.]

Edmon <edmon@neteka.com> wrote:

> Yes, I meant that when a dns application connects to the destination
> server, it includes the "original domain name".  Do you at this point
> use ACE or UTF8 or some other encoding?  IDNA mandates the use of
> ACE.  I think we should have an "option" where if the destination HTTP
> server is IDN/UTF8 aware, then the UTF8 could be used.

But "DNS applications" are the exception.  In the common case,
applications use the resolver, in which case the language spoken between
the resolver and DNS server is independent of the language spoken
between the application and the resolver.  (If a resolver provides a
UTF-8 interface but the DNS server does not support UTF-8, then the
resolver should ACE-encode the queries and ACE-decode the responses for
applications using the UTF-8 interface.)

Does IDNA really forbid DNS extensions for UTF-8?  Or does it merely
specify how IDNs can be used without extending DNS, while remaining
silent on the issue of DNS extensions?  I see no reason to forbid a DNS
extension for UTF-8, but nor do I see a pressing need for one.

> This would give the incentive for related servers and protocols such
> as HTTP servers to update to IDN/UTF8 extension.

I don't see how the incentive for other protocols to use UTF-8 is any
greater if DNS uses UTF-8.  DNS responses are seldom directly seen by
applications, and domain names obtained from DNS are seldom copied
into application messages.  The only time UTF-8 support in DNS would
encourage UTF-8 support in application protocols is when domain names
are copied directly from DNS responses into application messages, and
the frequency of that is seldom * seldom = approximately never.

A UTF-8 resolver interface, on the other hand, would indeed encourage
other protocols to support UTF-8, because then the applications could
copy domain names directly from application messages to the resolver and
vice-versa, without ever needing to do ACE encoding or decoding.

Mats Dufberg <dufberg@nic-se.se> wrote:

> The destinction between resolver and (authoritative) nameserver is not
> as sharp as might seems.

I wonder if you misunderstand what I mean by resolver.  The resolver is
a library on the client machine, linked into the application process.
It is not a DNS server of any kind--it is a DNS client.

AMC