Re: [idn] opting out of SC/TC equivalence

["Edmon" <edmon@neteka.com>]

John,

> And they should somehow to contrive to be sure that no records
> are cached outside China, where other rules would apply as a
> query is applied to a cached record?   I continue to be afraid
> that the choices are between
>
They do not have to contrive.  If a UserA types in <TC>.cn in Finland, the
.cn registry will do the TC=SC folding and respond saying that <TC>.cn=IP
and this will be cached in the Finland ISP resolver.  Now if a UserB types
in <TC>.cn, this could be responded using the cache.  If a User C types in
<SC>.cn, then the request will reach CNNIC and will resolve <SC>=IP and this
will be cached in the Finland ISP resolver.  This does not fragment the DNS.

> Now, if CNNIC says, "ok, we are going to register only SC, and to
> get TC translations, you need to use our translating software
> However, if you have client software that doesn't use our
> translation software and tables, you had better make queries in
> SC, and reverse mapping records are going to return SC no matter
> what you do", that would work with your model.  But I haven't
> heard from them going that route is an accceptable model.

The TC=SC mechanism could be done on the TLD server end.

> As I think others have pointed out, this works only if you assume
> that TC/SC equivalence can be solved by exactly two
> registrations, one of each.  As soon as there are combinations of
> the two forms, the combinatorics get you in major ways.

If you employ a server-end mapping mechanism, there would be only one
registration.

> If language issues are left to "implementations", we almost
> certainly end up with different implementations interpreting the
> same string in different ways and resolving it to different
> targets.  That is, in essence, the fragmentation problem.
>

I think I need to rephrase.  What I meant is that language issues should be
dealt with in a higher layer than the core DNS, that is implementation of
applications that will access the DNS.  Perhaps an application when prompted
with a particular name will perform multiple DNS requests to determine the
closest semantic representation.  If more than one is found, then it would
prompt the user to choose.  This is just a very very crude idea, but my
point is that linguistic/semantic issues should and could be dealt with in a
higher layer.

Edmon