Re: [idn] Comments on protocol drafts

[Paul Hoffman / IMC <phoffman@imc.org>]

At 07:01 AM 2/7/00 +0100, C C Magnus Gustavsson wrote:
>Well, 17 bytes is not a lot but it adds up. (Remember that it's not
>unlikely that it will be used frequently for decades or centuries.)

Quite true. However, as Randy points out, we've almost always been too 
conservative when trying to save bytes. Having longer name parts will have 
two negative effects on the DNS:
- It makes most resolution requests and some responses longer
- It requires more RAM in resolvers
Both of these are real considerations, but there doesn't seem to be much 
evidence that they will have significant impact. Of course, they are both 
easy to fix silently by adding pipe and RAM to resolvers as needed (just 
the same as happens all the time pre-idn).

>More importantly, it's completely unnecessary.

One of the requirements that most people have agreed on is to do the 
minimum damage to protocols that rely on DNS. Many of those have assumed 
ASCII. The word "completely" above seems a bit of an overstatement.

>Therefore, I propose the following requirement:
>
>   If an encoding is used, the ASCII characters in a string must not
>   be encoded in different ways depending on what other characters
>   the string contains.

If you are proposing this requirement for bandwidth/memory reasons, you may 
want to justify it with real numbers. If you are proposing it for aesthetic 
reasons, maybe it is a goal but not really a requirement.

FWIW, cidnuc is not the only text-based proposal that has been proposed; 
there's also UTF-5. In most cases, cidnuc will lead to shorter encoded 
strings than UTF-5; on the other hand, UTF-5 is simpler to implement and to 
explain. The only reason I proposed cidnuc was that I was concerned that 
some name parts that might be desired would not be expressible in UTF-5 due 
to length constraints. For non-European scripts (that is, the majority of 
the world), cidnuc encoding lengths are about the same as those UTF-8. 
Either way, it doesn't seem that encoding length is nearly as much of a 
technical issue as breaking DNS-reliant protocols.

--Paul Hoffman, Director
--Internet Mail Consortium