- From: Larry Masinter <masinter@parc.xerox.com>
- Date: Tue, 12 Nov 1996 08:52:49 PST
- To: w3c-dist-auth@w3.org
I woke up this morning with a strong feeling that we're baking too much into the URL syntax, and that while these kinds of decorations are interesting and important for implementors of versioning servers, there's no good reason why a client should be aware of them. Suppose that a versioning-aware server just returns a LINK header, e.g., GET http://www.cern.ch/TheBook/chapter2 HTTP/1.1 returns headers that include Link: <http://www.cern.ch/meta-repository/TheBook/chapter2>; rel="MetaData" and then the client will know that it can do GETs against the linked URL to get and modify the attributes of the included URL. This allows us to suggest a way of manging URLs in a particular server implementation, but, at the expense of another roundtrip, lets the clients be independent of the manging chosen, and gets us out of the debate about how one URL should best be mangled to produce another one. As for dealing with a single metadata URL (Rather than an attribute-at-a-time), I'd suggest that defining a media type (text/metadata?) as 'external package for document metadata' and return it all at once for retrieval, but have a separate set of operations (POST or UPDATE) methods for actually changing individual (or several) elements of the metadata. Regards, Larry
Received on Tuesday, 12 November 1996 11:53:02 UTC