RE: General internationalization

[Andrew Draper <ADRAPER@altera.com>]

> 2.2 Internationalization (I18N)
> 
>      Internationalized characters must be allowed to be represented and
used
>      in DNS names and records.
> This is good.
>      Implementation must specify what character
>      sets are used and how these characters are encoded in the DNS names
>      and records.
> This is bad. I strongly disagree that there is a *requirement* for
> multiple character sets or encodings. Those who feel that this is a
> requirement should say why it is so. I believe that adding 
> "All" to the
> beginning of the first sentence is sufficient.

Agree (with the modified first sentance).  Perhaps a better second sentance
would be:

If multiple character sets are used then the protocol must make it obvious
which character set is being used for each protocol element.  Alternatively
the protocol could use a single wide character set.
 
>      This document does not define any character sets that should be used
>      for I18N. However, non-standard character sets must not be used to
>      avoid duplicate work on general I18N. If multiple character sets are
>      used, they must be clearly identified.
> Same disagreement.

Agree. We should not presuppose that the protocol will use multiple
character sets.

>      The DNS protocol should remain deterministic. No DNS 
> element (resolver,
>      network, server or zonefile) should be required to do guess work.
> I do not see why "network" or "zonefile" is here; they are inherently
> not capable of doing guesswork. I'm not sure I even like 
> "server" here.
> I fully agree that resolvers should not have to do any guessing.

Agree that network should not be present.  The zonefile comment is badly
expressed.  What I meant it to mean was that if a server allows zonefiles to
be defined in multiple character sets then the character set of a zonefile
should be deterministic.  Of course, given a sufficiently intelligent server
zonefiles can be in multiple character sets independent of whether the wire
protocol uses a single character set or multiple character sets.

Is this requirement so obvious that it should be removed?


>      Must allow I18C in DNS RR response.
> I don't know what an "RR response" is.

Resource Record response.  I think this is intended to mean things like PTR
which return labels.

>      Must allow I18C in DNS TXT records.
> I don't think I agree with this since it has nothing to do 
> with domain names.

Not sure I agree with this either.  Many people are putting binary data into
TXT records - this requirement would break that.  Suggest instead:

"Must provide a record which can contain internationalised text (similar to
TXT)."

Of course whether to do this or not is entirely independent of putting
internationalised characters into DNS labels.

[more stuff snipped]

    Andy