Well, filename extension mappings are mostly a local problem. There
do need to be some additional suggestions.
One needed change is that the HTTP spec needs to at least hint at what
a server should do if a document is requested but the server cannot
deliver it in any of the Accept:ed formats. Return the document as
application/octet-stream? Return an error? Which one?
In any case, this also has problems with existing systems. Servers
could, in principle, assume that accepting "text/html" with no further
qualifiers means support for up to version 2.0, but handling "*/*" and
"*" (what the latter means is beyond me, but NetScape seems to
generate it) is much harder.
I don't think we're willing to declare that "When the client says it
will accept anything, it doesn't really mean *anything*" and require
every server implementation have complete historical knowledge of
client history. This is going to make it mighty difficult for servers
to allow the same document to be available in HTML 2.0 and HTML 3.0 in
a transparent fashion.
-- Marc VanHeyningen <http://www.cs.indiana.edu/hyplan/mvanheyn.html>