Re: [idn] The layers and character handling in DNS

[Patrik Fältström <paf@cisco.com>]

At 12.09 -0500 01-02-18, John C Klensin wrote:
>--On Sunday, 18 February, 2001 08:30 -0800 Patrik Fältström
><paf@cisco.com> wrote:
>
>  > > - prohibit some code points as part of a label (compare with
>>  the problems with having no definition of what is a hostname
>>  compared with definition in DNS protocol to be able to handle
>>  decimal values of 0-255 (inclusive) as part of a label).
>
>I think this takes us well out on a slippery slope.  And much
>further out if the rules/ prohibitions are not _extremely_ short
>and simple.

Yes, it is a slippery slope, but we already do the same thing when we 
use only ASCII in protocols. Now when having many more characters to 
pick from, we have to do the same kind of selection. I don't see how 
we suddenly can _not_ do it, and for example allow '/' (ascii slash) 
as a character which is allowed. I.e. we have to at least have the 
same ascii characters still being forbidden in textual based 
protocols in applications as we have today.

>  > - specify what should happen with non-assigned characters
>>  (which solves the versioning problem).
>
>Solves the versioning problem iff you believe that we will get it
>exactly right, for all currently-assigned characters, the first
>time.   Watching the Nameprep discussions over the last few
>months does not lead me to be optimistic about this.

Two different issues:

(a) UTC claims that assigned code points will not change. So, we 
either have to trust them or not. Yes, this is strong wording, but it 
was intentional.

(b) People in IETF don't understand that the IETF do _NOT_ do 
normalization. My job as liason (and some others) is to see that the 
IETF does not do anything which where we instead reference UTC.

So, some people bring up "bugs"/"issues" or whatever we should call 
what is brought up on the mailing list.

I will strongly oppose personally to any such mappings/fixes ending 
in any document from the IDN wg.

>  > - specify in what order case folding, normalization etc should
>>  happen.
>
>Yes, this I agree with.  But this one can be done as a few simple
>rules or not at all.  And we had better be able to enumerate
>"etc" and not need to add anything to that list in the future, or
>we will have another versioning problem.

Yes. I agree.

   paf