Re: [idn] Prohibit CDN code points

[John C Klensin <klensin@jck.com>]

--On Tuesday, 22 January, 2002 08:06 -0800 Dave Crocker
<dhc@dcrocker.net> wrote:

> In general, threads like the current one are examples of a
> failure to appreciate the complexity of creating special
> conditions and rules for Internet protocols.  Such
> idiosyncrasies in protocols make them much more difficult to
> implement correctly and usually destroy their ability to scale
> well.

Now it is my turn.  While Dave and I have a history of
disagreeing about many things, I appreciate his kind words about
my analysis of the equivalency problem.   And the above
summarizes, much better than I have been able to do, one of the
key problems we face in trying to make language-specific,
country-specific, or script-specific rules.  

The DNS, especially as represented by contemporary
implementations, is a fairly complex bit of protocol work.
Given that complexity, the code base, and some history of
fragile or incorrect setup of various domains, it is a
testimonial to the quality of the design that it works at all,
and does so without frequent major disruptions.  It is very hard
to imagine that we could change the protocols and implementation
to support a potentially-large collection of "if the TLD is X,
then..." and "if the TLD is Y and the SLD is Z" rules while
preserving the current level of robustness.   And, while adding
a Class would help isolate the current DNS from the problems
this would cause, it wouldn't make the ultimate complexity, or
risk, any lower.

We've heard a number of suggestions on this list by people who
have preceeded their remarks with words like "I am not an
implementor, but...".  This is the point --and the robustness
and scaling issues that Dave identifies are the key-- at which
looking at the problem and not at the protocol infrastructure
becomes a problem.  And so, I would strongly encourage those who
are suggesting "equivalence", or language-specific, or
script-specific systems to be sure they have recently read and
understood the basic DNS protocol standards and, preferably,
taken a careful look at the code for BIND or some equivalent
implementation.

     john