Re: [idn] Adding "optional" characters in draft-ietf-idn-nameprep

[Patrik Fältström <paf@cisco.com>]

At 21.25 +0000 00-08-13, Brian W. Spolarich wrote:
>On Fri, 11 Aug 2000, Paul Hoffman / IMC wrote:
>
>| a) We can ignore these characters on input (that is, toss them out of
>| the input stream).
>
>   This would be kind of ugly from an 'intuitiveness' standpoint.  Throwing
>away bytes that the user supplies seems like a bad idea, unless you can be
>fairly certain that this behaviour matches the expectations of the vast
>majority of 'reasonable users' (whatever that might mean).

I think this discussion is confusing because you mix how the user 
interface is to work with what the dns systems itself is doing.

Counter-example:

Is "foo" and "www.foo.com" two domainnames which are the same or not?

Both give back the same information to the user when typing them into 
a browser, and even the scheme "http" is default.

I feel this is what you talk about.

>| b) We can prohibit the characters on input.
>
>   Ack, I don't like this either.  How are we treated prohibited
>characters?  Stripping them out of the stream, or are we expecting the
>resolver to complain to the user?

I don't feel you can prohibit characters on input. People will write 
all different kind of applications. Some will handle different input 
scripts than others.

The important thing is to define what characters are prohibited in a 
"domainname".

I see the process be as follows:

    User enters a string.

    Application do something (undefined) and convert/extract what
    is supposed to be a domainname. This domainname is not allowed
    to include some characters (see Paul Nameprep draft), and
    normalization etc is also to be finished.

    This domainname is maybe encoded, and used for queries in DNS.

The second step is also divided in three sub-steps:

    Application do private things (like adding "www")

    Application do normalization

    Application do something with illegal characters

I don't care what the application does. Tell the user the string was 
illegal? Ignore the characters (remove them)?

The interesting thing for us is the output of these steps, and the 
output is not to include the forbidden characters.

>| c) We can allow them in names. This would mean that people
>| registering names would have to register them with and without the
>| characters (possibly in many combinations).
>
>   I'm not sure I see this as a requirement.  Why should the DNS be this
>smart?  If I put the string in the BIND zone file with the vowels, then I
>should expect to require them on the other end.  I think realistically
>users would at most store two versions of the label: one completely
>without the optional characters, one completely with.

Exactly. DNS should be stupid. It is today, and it should stay stupid.

>   My sense is that it is better to proscribe as little as is practical and
>let the user community establish their own conventions (e.g. the 'www'
>label for identifying web sites).

I agree.

   paf