Delay in discovery spec

[Eran Hammer-Lahav <eran@hueniverse.com>]

Recent discussions with members of the TAG, as well as other feedback
pointed out various issues with redirections and other HTTP restrictions in
my spec. There are also significant differences between how the three
methods are applied.

For example, consider the case where an HTTP GET for URI X returns a 301.
The Link header approach will follow the redirect and get the header for the
next URI in line. Site-meta cannot do that and will produce (at best), the
Link-header equivalent of the 301 header... This is not good.

In other words, the discovery spec must not decide which representation is
more appropriate (the one on the 301 reply or the follow-up 200 reply). This
requires the introduction of a new Context URI concept (actually introduced
in the upcoming Link header draft).

The client must first figure out the URI associated with the most
appropriate representation it is interested in. So web browser can choose
this to be the 200 response after following redirection, while a semantic
web application can use the 301 response (or both).

After the appropriate representation is identified, the URI dereferenced to
obtained it should be the input in the discovery methods, and follow up
request can be made (not following redirects). Whatever we get for that URI
is *it*.

None of this changes how we were actually going to implement all the use
cases we discussed, but it changes a lot how I need to explain it. This is
my top priority this week and I hope to ship -02 tomorrow or Wed.

EHL