1. something generic gets the XML and parses it
1.a the generic something displays the XML to a human using a
stylesheet, possibly with help from other logic for
embedded chunks
from different namespaces.
1.b the generic something invokes some application.
The 1.a example is of course Web browsers (although it seems there
are some nasty architectural holes in how you dispatch control for
embedded chunks of XML). I have a hard time thinking of a 1.b
example.
2. An app gets invoked to deal with a resource that happens to
be encoded in XML, and the app uses its built-in XML parser.
This case probably wants its own media type; the fact that
the encoding is XML is hardly material.
I would say that 1b is roughly equivalent to 2, except for an additional
level of abstraction.
I actually prototyped, a while ago now, 1b using the standard set of
mail tools on Unix. I basically specified a compound document/application
object in a TR9401 catalog (a poxy medium it turns out, that needed a lot
of fixing, and a number of extensions). That catalog could be transmitted
to a generical application that then invoked whatever else was appropriate
(in this case, DynaText, Panorama, and Web browsers).
I think we need to clearly seperate the issue of packaging a single XML
instance/entity, and a compound document. I would argue that the former
is almost entirely devoid of any application semantics (being either
simple text/xml or application/xml), but the latter *intrinsically*
has *some* semantics associated with it.
I guess this is basically supporting the notion that application specific
uses of XML should have application defined media types, but there is a
slight difference in the *mechanism* that I am talking about. We could
have something like xml/processing-specification or something suchlike,
with an associated handler, and that handler could be reponsible for the
dispatch (not part of the MIME world at all). Again, I prototyped something
like this some time ago, and it did work well, even on top of *existing*
(this was '95) MIME infrastructure.
I should note that both Simon St. Laurent and Chris Lilley have been making
noise about such "catalogs" recently as well.