Re: [idn] Canonicalization: [28] through [31]

[RJ Atkinson <rja@inet.org>]

At 11:52 26/06/00 , Paul Hoffman / IMC wrote:
>Not really. The requirements document has the following (editied for clarity):
>
>+---------------+
>| Application   |
>+---------------+
>       |  Application service interface
>       |  For ex. GethostbyXXXX interface
>+---------------+
>| Resolver      |
>+---------------+
>       |     <-----   DNS service interface
>+-------------------------------------------+
>
>"Before the DNS service interface" could mean "in the application" or "in the resolver". I do not believe that this WG has come to consensus on this. Thus, I very carefully worded my proposed sentence to say that the canonicalization happens before the DNS service interface, but to limit which of the two places it might be. The protocol must pick one or the other, but it is too early to say which.

         Putting it in the application is inherently broken because numerous
existing deployed applications that aren't going to change anytime soon 
and will remain deployed do not perform canonicalisation at present.
Relying on software known not to perform i18n canonicalisation to perform
canonicalisation leads to chaos and can't be interoperable.

         So I do not believe that requiring canonicalisation be performed
by applications is technically a valid choice.  I believe the only
two technically feasible locations are inside the DNS client resolver
or in the DNS server (including caching servers).  

If folks want to disagree with my conclusion, please explain in detail
how interoperability is maintained in the presence of these existing 
applications that do not perform canonicalisation and won't change 
any time soon (if ever; some are binary applications that are deployed
but do not have any active development and do not have source code
available).

Ran
rja@inet.org