ISSUE: URI form for hash draft

The current hash draft allows a URI to be on either the left hand side
or the right hand side of feature declaration.  I have proposed to the
authors that they consider limiting the URI to the right hand side, so
that we do not imply an architecture for dereferencing the URIs that
we do not provide.  I also believe that limiting things to the hash
form will make it easier to write and maintain parsers, especially
in the presence of proxies.  

Graham and Larry have raised a major issue with this proposal,
however, and I believe the group needs to consider it carefully as it
has a variety of implications for the use of the CONNEG framework.

The issue is that limiting the URI to the right hand side potentially
adds either bytes or round trips when dereferencing occurs.  It adds
bytes when protocols include a hash :- URI line with the original
reference.  It could add a round trip if there were a hash sent that
the recipient did not understand, then a dereference to the URI, then
a second dereference to the feature set if the URI is not understood.

For those who have not reviewed the draft yet, please do so as soon
as possible, with an eye to this problem; if you have reviewed the
draft, your comments on this issue would be welcomed by me and
the authors.  In particular, applications of the hash function to
specific protocols using the CONNEG framework are solicited.

			regards,
				Ted Hardie