- From: Roy T. Fielding <fielding@gbiv.com>
- Date: Wed, 8 Sep 2004 17:18:28 -0700
- To: Sandro Hawke <sandro@w3.org>
- Cc: www-tag@w3.org, Mark Baker <distobj@acm.org>
On Sep 8, 2004, at 3:11 PM, Sandro Hawke wrote: >> Hmm, I think this paragraph goes too far ... >> >> On Wed, Sep 08, 2004 at 04:30:13PM -0400, Sandro Hawke wrote: >>> Information Resources are the only kind of resources which can >>> have >>> representations. The number 1, which is not an Information >>> Resource, might be said to be represented by the two-octet >>> sequence >>> 0x0001, but not in the sense of "representation" used in this >>> document. >> >> No? Why not? It seems to me that it does. > > What information is contained in the number 1? All that is inherent in its identity as a concept. If that were not information, the concept could not be taught to math students, let alone small children who just want to know how many helpings of dessert they are allowed to eat. Since you can communicate the concept to me with just two words (and a lot of context), the concept can and is represented as such even if that is not sufficient (alone) to encompass its "meaning". >>> A web-accessible control dial, set to "1", might respond >>> to HTTP GET requests with a representation of its state: 0x0001. >>> In this example, 0x0001 acts an identifier for the number 1 within >>> the data format being used. No, it acts as one representation of its state as mapped to the format described by the media type. It doesn't say anything about how the dial is implemented, nor the meaning of 0x0001, though that may be discovered elsewhere or elaborated by the media type. >> I think it's representing, not identifying. If it were an >> identifier, it should be a URI (or an EPR, I suppose 8-). > > It could indeed be a URI, but people (eg Patrick Stickler in RDF Core > :-) argued that more compact text/bits were needed for things like > numbers. Umm, like "n:1"? [I don't believe that all identifiers need to be URIs]. Let's say that, instead, the response comes back as HTTP/1.1 200 OK content-type: application/uri urn:integer:1 (and someone actually bothered to register those things) where the media type is defined to mean "that which is identified by the enclosed URI". What then? Is that not capable of representing *everything* we know and mean by the abstract concept of 1? Furthermore, any degraded form of representation is also sufficient to represent 1 provided that the recipient can still understand its meaning, in the same way that various lossy graphic formats can be used to represent the same resource as a 24bit TIFF. That is what representations do: they do not define the nature of the resource -- they merely have to be consistent with the resource. >>> An HTTP GET of a URI for the number 1 >>> itself could meaningfully be met with an error or redirect, but >>> not >>> with a representation. >> >> Gotta disagree with you there (even before pointing out that errors >> and redirects are representations 8-). > > The MIME Entity (representation) carried in an error or redirect HTTP > response (in the cases where there is one, like 404) is not stated in > the spec as being a representation of the named resource, as far as I > can tell. It is a representation of the error as defined by the server, not of the requested resource. Mark is right -- the assertion you give does not correspond to how representations are used on the Web. Personally, I think the distinction between resource and information resource is unnecessary. It is an attempt to fix broken assertions by claiming that there is an artificial divide among classes of resources, thereby defining systems that bridge that divide to be "bad" just to satisfy a philosophical argument over the limitations of RDF. I think it would be better to simply accept that some assertions are on resources and other assertions are on representations of resources, and any attempt to make an assertion on a resource that isn't true of all of its representations is just a bad assertion (like any other bad assertion). It can be fixed by defining new property URIs that are more specific about what is being asserted, or by doing what TopicMaps does in distinguishing the targets. It cannot be fixed by creating an artificial divide, because the same problems occur when making assertions about a content negotiated (information) resource. And the only reason this argument has persisted for so long is because people insist on framing it as something "new" that only affects the semantic web, and using obscure examples of dogs and cars. If you simply take the same example and apply it to "my home page", wherein the page actually consists of multiple representations that are independently authored and may or may not be provided depending on the requesting user agent, then you can use the same invalid dc properties to make contradictory assertions about "my home page" that you would for a "car" or "picture of the car", or for anything else for that matter. This is not the fault of the person who had the nerve to place "my home page" on the Web: it is the fault of using ambiguously targeted assertions within a language that assumes properties are unambiguous. ....Roy
Received on Thursday, 9 September 2004 00:18:38 UTC