Re: idn-uri document

Hello Erik,

Sorry to be late with my answer. I started writing it when
I got your mail, but it somehow got lost.

At 19:09 02/09/13 +0200, Erik Nordmark wrote:

>Martin,
>
>I just read draft-ietf-idn-iri-01 and I have some comments and
>questions.

Thanks for your interest!

>I don't know if the best place to discuss this is the IDN list, or
>if there is some other list looking at URI/IRI stuff.

For URIs, there is the old (IETF WG) uri list, now hosted at W3C:
[email protected]. That would be appropriate because the draft is currently
written as an update to the URI spec. Roy Fielding is working on
a new edition of RFC 2396, mostly bug fixing.

The W3C Internationalization WG is looking after the IRI draft.
See e.g. http://www.w3.org/International/iri-edit/. So they
could take it on. Public discussion would be on [email protected].

The IDN WG is of course the third alternative, that's where it's
currently, although it hasn't really been discussed a lot.


>On page 4 it says "will always be rejected by resolvers".
>I don't know if this is intended to be a statement about the current
>implementations of resolvers, or a statement about something we should
>recommend or require resolvers to do.

I added an explanatory sentente 'because no such domains will be
registered'. So it is about current (and hopefully future)
operations, not about dns server implementations or resolver
implementations.


>I do think there currently are resolvers which happily pass whatever
>string of octets into DNS packets and send them off.

Ah, I see, of course the resolvers pass things through and then
pass the negative result back, so they don't actually reject it.
So now the sentence reads:

However, such syntax should never be used, and will never be
resolved because no such domains will be registered.


>And I'm far from certain it would be a good idea recommend or mandate
>that resolvers do additional checks. The IDNA model is that the clients
>do nameprep and that the DNS servers just to a (ASCII case insensitive)
>exact match.

Agreed, sorry about the confusion.


>The defined syntax rules for declare certain ASCII domain names illegal
>(such as *.example.org). Where is the check for illedgal names assumed to
>be performed? For IDNA it probably makes sense to only apply this types
>of checks (setting the UseSTD3ASCIIRules flag) when verifying domain name
>registrations and not do such checks in the clients.

This is an IDNA question, not a idn-uri question. As far as I remember,
the idea was to have the checks done on the clients, too (with some
leeway for unassigned characters to stay forward-compatible with
new character assignements). The reason for this was to create
pressure on registries to follow the rules.


>      The work of the IDN WG includes some procedures for name preparation
>     [Nameprep].  Before encoding an internationalized domain name in an
>     URI, this preparation step SHOULD be applied.  However, the URI
>     resolver MUST also apply any steps required as part of domain name
>     resolution by [IDNA].
>
>The above statement says that for all domain names (note that the term
>"IDN" is defined to include the existing ASCII domain names)
>one should apply nameprep. This might be fine but it makes sense
>stating this explicitly. The ToASCII in IDNA does not apply nameprep
>to all-ASCII labels.

The idea was simply to say: We RECOMMEND that you apply the IDNA rules
already when you create an URI. What these rules are is up to IDNA.
If IDNA says that their preparation of ascii-only labels is the
identity operation, then we recommend that you apply that (i.e. do
nothing), and not something else. If you see a way to make this clearer,
please tell me.



>Always applying nameprep will have the effect
>of downcasing the ASCII characters in all ASCII labels, which IDNA does not
>do.

And of course we don't want to do either.

>Which are the "any steps required as part of domain name resolution"
>above? I can't figure out to what it might refer.

That's the nameprep and related checking that the client has to
do when it resolves a domain name. In IDNA terms, 'client' would
be easy to understand. But using the word 'client' in an URI context
doesn't work, so I tried to word around it. Any improved wording
appreciated.


>Finally, is the intent that nameprep always be applied before characters
>are encoded in UTF-8? Then it makes sense stating that in the first real
>paragraph on page 4.

No. In the context e.g. of IRIs, the conversion from an IRI to an URI
would not do nameprep.

Regards,    Martin.