Re: IRIs, IDNAbis, and HTTP [i74]

On 14/03/2008, at 9:26 PM, Julian Reschke wrote:

> Mark Nottingham wrote:
>> Personally, I am *very* -1 on doing this.
>> Changing the allowable characters in a protocol element is a *big*  
>> change, and there is not an interoperability gain to doing so.
>> There is also not a functionality gain; it is possible (if not  
>> pretty) to serialise other characters into HTTP headers.
>
> I think the key question here: is that implemented in practice? (In  
> particular, which encoding?) If yes, fine (and maybe let's document  
> what works). But if not...?
>
>> Furthermore, HTTP headers for the most part don't carry user- 
>> visible data, and when they do, it's often an IRI, for which there  
>> is a well-understood serialisation into header-safe characters.
>
> None of the usages in RFC2616 is a IRI (by definition :-).
>
> And, as Brian observed, IRI->URI mapping is not sufficient if you  
> need to *exactly* reconstruct the original IRI (such as when the IRI  
> is used as a name, e.g. an Atom link relation).

I think Larry was making much the same point at the app area arch ws,  
IIRC.

My gut feeling on issue 74 is that if any encoding (e.g., IRI->URI,  
RFC2047) is usable in a header, it should be specified on a header-by- 
header basis explicitly, not inherited as a general rule like this.  
That allows the flexibility to use the right mechanism for the job,  
and keeps the overhead low.

Are there any other headers (besides URI-based headers) that we need  
to consider, beyond these?
  * WWW-Authenticate, Proxy-Authenticate, Authorization, Proxy- 
Authorization
  * Content-Disposition
  * From
  * Warning
  * PICS-Label

--
Mark Nottingham     http://www.mnot.net/

Received on Friday, 14 March 2008 10:47:35 UTC