- From: Roy T. Fielding <fielding@gbiv.com>
- Date: Tue, 28 Mar 2006 10:54:46 -0800
- To: noah_mendelsohn@us.ibm.com
- Cc: W3C TAG <www-tag@w3.org>
On Mar 28, 2006, at 7:05 AM, noah_mendelsohn@us.ibm.com wrote: > last week we decided to have a more balanced presentation of the > upsides > and downsides of relying on inferred metadata. In particular, we all > regularly manipulate URIs based on guesses involving metadata. For > example, after somehow discovering a link to: > > http://example.org/book/chapter2 > > many of us will quite reasonably experiment with trying to find > chapter 3 > at: > > http://example.org/book/chapter3 > > even if we haven't checked what the publishers actual URI assignment > policy is. If the document retrieved by that URI looks right, we may > trust it. Of course, you wouldn't use such a guess when requiring > information that's 100% reliable, but in many cases it's a good an > useful > thing to do. It's an important aspect of what people expect of > the Web. Umm, yes, but that isn't metadata about the resource. The 2 and 3 are essential parts of the identifier. The finding should be about non-identifying data included in the URI, which is almost always stuff that is metadata about actions on that resource. Hmm, I know I've sent this message before .... oh, it was private mail. Here is what I wrote on May 23, 2003: Message-Id: <ECAFA7D8-8D64-11D7-A116-000393753936@apache.org> I think it needs more work. We need to distinguish between metadata and identifying information. Versioning, for example, is identifying information and does belong in the URI iff the resource is that specific version. The reason we rejected the request was not because it isn't identifying information -- it is because there is no need for a single standard on how the origin identifies versions. Metadata, OTOH, is ancilliary information -- what is identified does not change with or without the metadata. As such, its presence causes the number of possible identifiers per resource to explode. Having too many identifiers for the same [resource] causes all sorts of havoc with caching, access control, etc. Finally, the opacity principle is not saying that servers should not include meaningful information in the URI -- it is saying that clients should not derive meaning by inspection of the URI. Note, however, that the opacity principle is scheme-dependent -- it is not true for ftp and file schemes, among others. I'd suggest more, but I really need to finish the URI specification, particularly since this finding is dependent upon that. ....Roy
Received on Tuesday, 28 March 2006 18:55:16 UTC