RE: [idn] Unicode tagging

[Andrew Draper <ADRAPER@altera.com>]

Keith Moore wrote:
> I'm not sure why SOA records should be a big issue in any case.
> If email addresses in SOA records remain in ASCII, this doesn't
> seem like a big problem - since they exist for operational purposes
> (where having globally-usable addresses seems like a huge advantage)
> and they are not normally seen by users.

In fact SOA records are likely to be easy since they appear only in the
RDATA (on the right hand side) and so don't need to be compared (so
canonicalisation and downcasing will only be used for optimisations such as
name compression).

The harder case is when the owner name of a RRSET (what appears on the left
hand side, and which does need to be matched) is not a hostname.  Two
examples of this are the RP record and the KEY record when the appropriate
bits in the flags field are set (see RFC 2065 section 3.2).  The KEY record
is particularly difficult since the least significant (leftmost) label in
its owner name is either a hostname or an email address LHS depending on the
contents of the RDATA.

Given that such labels (parts of owner names which are not hostnames) exist
we must work out what to do with them.  Saying that they are "not a big
issue" and ignoring these names seems irresponsible.

I suggested four options earlier:

1) Require that non-hostname labels remain in ASCII only.
2) Require that protocols allow non-hostname labels to be internationalised
with the same canonicalisation etc. algorithms.
3) Require that protocols allow non-hostname labels to be internationalised
with different (less restrictive?) canonicalisation etc. algorithms.

4) Require that the protocol must specify which of these options (or which
variation on one of these options) it has followed.

Which do people prefer? 

Regards,

    Andy