On Nov 9, 2004, at 12:35 PM, Tim Bray wrote:
http://www.intertwingly.net/wiki/pie/PaceFieldingLinks
I think it would be better if my name wasn't in the page title. I could come up with a dozen or so different ways of providing extensible links in XML -- this one was just the simplest way of making an attribute value extensible. I would not call any of them the "Fielding" way.
It would be simpler to modify the proposal to just say that the attribute is either a URI or a token, where the token is equivalent to the URI obtained from appending the token to "http://www.iana.org/assignments/relations/";. In fact, I'll do that right now.
Another alternative would be to remove the generic link construct and simply use namespaces for all of the metadata. Either XLink or RDF could be used to indicate when an attribute value is a URI reference (and thus a link). Note, however, that neither RDF nor XLink are simple, nor are they proven solutions for extensibility by any reasonable measure -- they are simply a means of associating crosscutting behavior (similar to aspect-oriented programming) as an attempt to recover a centralized meaning from namespaces' decentralized collision avoidance strategy.
None of that discussion will help the WG reach rough consensus, however. There are even valid arguments for not allowing any sort of extensibility beyond that of ignore-unknown-elements.
Tim, you might want to consider taking all of the link and metadata proposals and organizing them as a ladder, with levels of extensibility being the rungs. Then get each person to indicate how far up they are willing to climb. I'd do that myself, but I still haven't figured out why I subscribed to the list in the first place, aside from the pure entertainment value of watching a group of professional and semi-professional writers try to reach consensus on a design, and might disappear if I come to my senses.
....Roy
