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

[James Seng <jseng@pobox.org.sg>]

Lets not go to the detail of the design. We are working on the requirements.

Depending on the canonicalization we adopted, the process may not be too CPU
intensive that doing it several times may be 'okay'. I mean, heck, we already
doing strcasecmp() many times...

-James Seng

Patrik Fältström wrote:
> 
> At 23.01 -0400 00-06-26, RJ Atkinson wrote:
> >"Each proposal MUST clearly specify precisely where canonicalisation
> >is performed (e.g. DNS resolver, DNS server)."
> >
> >          If we don't make the above clear, then the deployed result
> >          just won't be interoperable.
> 
> I think we have had enough arguments by Paul et.al. stating that
> doing this in the server "is a bad thing".
> 
> But, I might be convinced that the text above is ok.
> 
> Note that IF this is done in the server, then the canonicalization
> for one label in a DNS query will be canonicalized several times,
> once for each server the label hits.
> 
> Given a query for foo.bar.example.se.
> 
> (1) Once in the local resolving server
> (2) Once in the root nameserver
> (3) Once in the server for se
> (4) Once in the serevr for example.se.
> (5) Once in the server for bar.example.se.
>      Ooooops....this was a CNAME...
> (6) ...
> (7) ...
> 
> This definitely should be done in the client!
> 
>     paf