- From: Patrick Stickler <patrick.stickler@nokia.com>
- Date: Fri, 05 Jul 2002 14:02:13 +0300
- To: ext Paul Prescod <paul@prescod.net>, Joshua Allen <joshuaa@microsoft.com>, WWW TAG <www-tag@w3.org>
On 2002-07-05 10:27, "ext Paul Prescod" <paul@prescod.net> wrote: > > Joshua Allen wrote: >> >> ... >> >> Maybe that is because the discussion has veered off-topic. There is no >> question that HTTP returns a representation of a resource. However, >> that does not mean that URIs using the http: scheme should be used for >> things other than web pages. > > There is nothing in HTTP specialized for "web pages" and many people use > it for things completely unlike Web Pages. Microsoft, in particular, is > a leader in using it for e.g. email and travel reservations. Just because someone does something does not make it correct or optimal. And as I've said many times before, there are things that are done that work fine for the Web but won't work for the Semantic Web because the latter has much higher demands for precision of denotation. >> ... HTTP is just one protocol, with semantics >> that are useful for a finite set of circumstances. URIs in practice >> utilize many schemes besides just http:. Saying that all resources can >> or should be identified with http: URIs is completely insane. > > It is indisputably true that all resources CAN be identified with http > URIs. After all an identifier is just a string and the protocol does not > effect the binding of identifiers to abstract objects. Whether > everything "should" be identified with http: URIs is a debate worth > having. Even without going that far we can observe that HTTP URIs have > the nice property that it is extremely easy for someone with a web > server and the appropriate domain name to scalably and globally > associate metadata in the form of representations with resources. Whoa, hold on now. You're now saying that metadata describing a resource is a representation of that resource?! And just how are we going to differentiate, say, between a representation of an RDF/XML instance which is bit-for-bit identitical with the actual resource and another RDF/XML instance which describes the resource? It seems the real problem here is the implicit overloading of URIs by the Web architecture which is far too low a resolution for Semantic Web agents. Yes, in theory, the Web architecture provides mechanisms to name each variant representation separately -- and thus, one could then describe each variant to express its characteristics and relation to the actual resource -- but that is seldom used in practice, and certainly does not appear to be encouraged by the Web architecture. The problem I see is that in many if not most cases, when the resource in question is digital, folks expect HTTP to return a bit-for-bit identitical copy of the resource. Yes, content negotiation can cause something else to be returned, but in the case of schemas, invoices, certificates, etc. and many other kinds of digital resources, we expect to get what we ask for. The fact that there is no distinction between a bit-for-bit copy of a digital resource and something else that is not the resource is the problem. If HTTP wants to return a representation, even using the very broad interpretation of representation such that a textual description of the city of Paris is treated as a representation of Paris, fine, but it should be consistently clear in the Web architecture itself, such as in the response codes from an HTTP server, that what is being returned is *not* a copy of the resource "at full fidelity" and should be considered as some other resource that is related in some manner to the resource denoted by the URI provided to the GET request. And ideally, the returned resource should *always* be named if it is not the same resource denoted by the URI of the GET request. Only then will semantic web agents be able to deal with the ambiguity of HTTP GET results. > urn: > URIs do not have that nice property and in my mind are thus strictly > inferior. No, given DDDS, urn: and http: URIs are no different whatsoever in this regard. Yes, http: URIs are easier to deploy at the moment, but they do not differ in the way you suggest, technically. Patrick -- Patrick Stickler Phone: +358 50 483 9453 Senior Research Scientist Fax: +358 7180 35409 Nokia Research Center Email: patrick.stickler@nokia.com
Received on Friday, 5 July 2002 07:02:08 UTC