On 24 Apr 2007, at 16:57, John D. Mitchell wrote:
This takes us, full circle, back to the confusion that started this thread originally... The implementation, and all of the consequential edge cases, lead to a confusing mental model of the expected behavior. This specific example of adding a "sub" reference to a "http://some.com/dir"; gives the erroneous answer of "http://some.com/sub"; rather than "http://some.com/dir/sub";. As you noted, if the directory is "always" (re)written to the proper, normalized form "http://some.com/dir/"; then the expected behavior is kept. Great but that doesn't lessen the confusion for developers who have to keep this extraneous complicatification in mind and make sure their code works regardless.
How do you suppose a client to know if a relative link in a representation retrieved from http://some.com/dir to "sub" is to "http://some.com/sub"; or "http://some.com/dir/sub";? You would need some new magical headers (WebDAV-ish) to say that the resource is a collection or you need to hard-code this knowledge in the client (which would make it harder to change http://some.com/someresource into a collection later when you want to add http://some.com/ someresource/morestuff)
I don't see a big point in exploring how this could have been done differently, URLs have existed for a long time already and I believe you need a better reason than "I don't want to generate the right URI to the collection (ie. with trailing /) in the first place!" to go outside the expected and normal behaviour.
-- Stian Soiland, myGrid team School of Computer Science The University of Manchester http://www.cs.man.ac.uk/~ssoiland/
