- From: Jim Whitehead <ejw@ics.uci.edu>
- Date: Fri, 13 Nov 1998 09:29:47 -0800
- To: "Slein, Judith A" <JSlein@crt.xerox.com>, WEBDAV WG <w3c-dist-auth@w3.org>
> Collections are resources, and their members are resources, and membership > in a collection has nothing to do with what identifiers are used for the > collection or its members. OK, but from a protocol standpoint, there is no way to talk about resources without using identifiers (URLs). As a result, I cannot see how to meaningfully discuss the membership of a collection without using URLs. > > Methods are defined for creating and deleting collections, adding and > removing collection members, copying and moving collections, listing > collection members, and listing the collections a resource belongs to. > (Only the last of these is missing today.) So we never have to > rely on the > syntax of identifiers for any of these operations. > > MOVE resource R from collection C1 to collection C2 just means if you do a > PROPFIND on C1, you will no longer see R in the result set, and But PROPFIND returns a URL, identifying the resource on which properties are defined. The requirement would have to be written "any URL which identifies R will no longer be seen in the result set". > if you do a > PROPFIND on C2, you will see R in the result set. It implies > nothing about > identifiers for R. > > DELETE collection C1 means that C1 and all of its member > resources are gone, > not just that certain identifiers for those resources don't work any more. > > Eventually we need to ask, can there be collections that weren't created > with MKCOL? In particular, may the file system directory hierarchies that > many Web servers expose be collections? We want to say, yes. All that's > necessary is that the Web server support PROPFIND, etc., for those > resources. May a server decide that some of the resources in the file > system directories are collection members and others are not? Yes, that > might be confusing to users, but it could be done. We expect that mostly > servers would decide to have their URL namespaces exactly mirror > collection hierarchies, so that mostly it would be possible to deduce collection > membership from the URL syntax, but no one should ever rely on this. OK, to make sure I understand, this is an argument against strict consistency for DAV resources, right? - Jim
Received on Friday, 13 November 1998 12:34:49 UTC