Re: CURIEs: A proposal

On Jun 26, 2006, at 8:57 AM, noah_mendelsohn@us.ibm.com wrote:
> Roy Fielding writes:
>
>> All URIs are dereferenceable, not just http URIs -- some
>> schemes are less amenable to that than others, but there is no
>> reason to say "http" here.
>
> Well, I meant to be silent on the status of non-http schemes.  Clearly
> there are many other schemes that are typically deferfenceable, and I
> think I understand the sense in which you mean that all are at  
> least in
> principle.

It is an occasional practice for some people to claim that they need a
new URI scheme because they *do not* want the URI to be dereferenceable.
That is a common fallacy.  All URI schemes become dereferenceable as  
soon
as someone can map any representation associated with the resource to a
mechanism that accepts a string and returns a representation.  It is  
just
harder than using the existing infrastructure for "http" (or "ftp",  
etc.).
If you want to be silent on the status of non-http schemes, then  
don't use
"http" in the sentence -- the implication is that "http" is somehow
different in that respect than other schemes.

> In practice, there is much less deployed infrastructure for
> dereferencing some other schemes, such as urn, and I didn't want to  
> open
> that side debate here, which I think might have happened if I had
> suggested that representations should be deployed for all namespaces,
> regardless of scheme.  I do agree that it's a good thing to do, not  
> just
> for http-scheme URIs, but whenever practical.  Turning the argument
> around, the widespread avaialbility of infrastructure for manipulating
> http resources is among the reasons for encouraging use of http  
> URIs as
> names for namespaces.

That doesn't change the principle that namespace names should be
dereferenceable *because* Web architectural knowledge should be
grounded in the Web.  You have misplaced the principle by trying to
avoid reference to other schemes.  The principle was intended to cover
all namespace names, not just those beginning in "http".

By limiting discussion to "http" namespaces, you are implicitly
suggesting to people that it is okay to use non-dereferenceable
namespace names provided that they are not "http" identifiers.
That is wrong. The principle must be stated in the general case
even if you happen to be focusing on "http" schemes at the moment.

....Roy

Received on Monday, 26 June 2006 21:48:03 UTC