- From: Xiaoshu Wang <wangxiao@musc.edu>
- Date: Sat, 27 Jun 2009 17:04:42 -0400
- To: Erik Wilde <dret@berkeley.edu>
- CC: Dan Brickley <danbri@danbri.org>, Larry Masinter <LMM@acm.org>, "'Pat Hayes'" <phayes@ihmc.us>, "'Eran Hammer-Lahav'" <eran@hueniverse.com>, "'Dan Connolly'" <connolly@w3.org>, "apps-discuss@ietf.org" <apps-discuss@ietf.org>, "www-tag@w3.org" <www-tag@w3.org>, "'URI'" <uri@w3.org>
Erik Wilde wrote: > hello. > > Xiaoshu Wang wrote: > >> There should not be. I have trying this many times. A URI, fragmented >> or not, denotes one thing and its returned representations another. The >> former is the content of the later. The remedy is to define a URI >> syntax for representation. >> A syntax that I have proposed is to insert a (mime-type) after the # sign. >> Thus, >> "http://danbri.org/foaf.rdf#danbri" denotes a person. >> "http://danbri.org/foaf.rdf#(application/rdf+xml)danbri" denotes an RDF >> node. >> "http://danbri.org/foaf.rdf#(application/xhtml+xml)danbri" denotes an >> HTML element ided "danbri >> > > interesting. would that be specific for http-identified resources? if > not, how would that be supposed to work with URI schemes that do not > share HTTP's capabilities for transferring content metadata, and > performing content negotiation? a simple example might be FTP, which is > similar in nature to HTTP (access to hierarchically organized resources) > but has no concept of media types. > I don't think a URI scheme has to do anything with transportation protocol. No matter what URI you use, after de-reference, you get a *representation*, which is a different thing from the *resource* that the URI denotes. And a representation must have a content type, regardless how you retrieved it. > another thing i am wondering about: aren't fragment identifiers as they > are currently defined client-side only and specific to the media type > anyway? that might indicate you are talking not about extending the URI > syntax, but that of HTTP URI fragment identifiers? See the above answer. > there were other > approaches of doing this (with other goals), one of the issues was how > to create some framework for fragment identifiers that would be > uniformly applied to all fragment identifier syntaxes. I am not sure what is the purpose for that -- a standard syntax for fragment identifier? It is just a name, right? What I am proposing is to make a syntactic notation on URI syntax so that we will no longer be pondered by the question of what a URI denote. > that one never > got anywhere, and the window of opportunity is probably closed by now. > but there the idea was that instead of labeling fragments with the media > type to which they should be applied (which seems to be what you're > suggesting), they should follow some base syntax, and thus could be > designed to be less brittle across media types. HTTP's idea of content > negotiation (and thus dynamic media type assignment at access time) and > URI fragment identifiers and their media type specificity always was one > of the areas where web architecture certainly could need a bit of > improvment. > Content negotiation does not cause the problem. It only makes the problem obvious. The Web is based on three fundamental entities: URI, Resource, Representation. But currently, the referential range of the URI only covers resource but URI and Representation. What I have proposed in my manuscript to ISWC 2009 is as follows. 1. If a URI's root is ended with a "~", it denotes the URI sans the "~". 2. Insert (mime-type) after # to denote a particular type of representation retrieved from a URI. 3. If a URI's root is ended with a "?", it denotes the list of all mime-types supported by the root URI. With #1, we can built a URI that is composed of any numbers of URI but yet still maintain a level of curtness because the mapping can be described in a representation and can be retrieved. This will solve most, if not all, problems raised in XSD use cases. The reason for #2 is for httpRange-14 that has hunted us for 7+ years. As a side note to Dan, I think mime-type should be formulated in URI just like any other resource. I have detailed my reasoning from my manuscript. One of the use case is that I have one resource but there are two available XML schema, which overlaps but neither one consumes the other. With the current media-type specification, it forces me to choose one, which is not what the best for me and for my potential clients. Also, there are other advantages of extending mime-type to URI. For instance, in principle, a client can "follow the nose" to retrieve something and parse a novel format. The #3 is in essence a transparent content negotiation. But modeling it as a kind of resource allows all kind of formats be used to describe the list. Also, at the most fundamental level, a mime-type is semantically equivalent to a service, so the #3 can be considered to be a default standard mechanism for service discovery. This thus essentially solves all the so-called "uniform access to metadata" problem. This URI pattern (TIP, I called it a pattern for now since it is not a in URI spec) + The And Pattern (TAP i.e., to supply a resource with one mime-type And another And another via content negotiation), gives us both diversity (in presentation) and uniformness (in organization and therefore discovery). Personally, I think that is all we need at the most basic architectural level of the Web. Xiaoshu
Received on Saturday, 27 June 2009 21:05:25 UTC