Re: Comments on the architecture doc

Mark Baker wrote:
> 
>...
> 
> If it is XSLT, what's the problem with wrapping it with an XSLT
> container, or with delivering it as application/xslt+xml?  Why
> insist (if I understand your position) that it be able to be
> delivered as application/xml *and* still be processed as XSLT?

I wasn't really talking about how this maps into the MIME system at all.
I was just pointing out that we are going down the wrong path if we try
to associate machine readable specifications with namespaces and then
apply them to parts of the document tree dynamically. Namespaces (as
defined today) do not represent useful "languages" or "document types"
as Tim and Tim implied.

To put it another way, the (URI, localname) pair is opaque, just as URIs
are opaque. You would never try to associate processing with URIs that
began with "www.*" instead of "ftp.*". So why does it make sense to try
and associate processing with (http://www.w3.org/1999/xhtml, *)? It's
the same deal. 

We could update the namespaces spec (literally or figuratively) to make
the root namespace magical and make namespaces non-opaque but that isn't
how it is today.

>...
> That's my point too.  The parent container(s) define the meaning of
> containment.  In this context, those containers are, in order; media
> type, root namespace, other nested namespaces.
> 
> In other words, the media type determines the way in which the root
> namespace is interpreted.  application/xml doesn't define that meaning.
> But that doesn't prevent us from defining another media type that is
> unambiguous about how root namespaces are used to trigger processors.

I don't think it will be clear to people how to use all of these
different namespaces. I would rather "just" use MIME types and perhaps
define an in-document analog, which we could call "inline MIME type" or
"document type".

XML needs to have well-defined rules. If the "new rule" is that the root
namespace is magical then XSLT should just be changed to conform. It
would be very confusing if the root namespace was sometimes magical. You
would save a document to your hard disk and find that it behaved
differently than when it was on the Web...unless there was a new file
extension. Which gets even more confusing.

> My soon-to-be I-D on this topic is;
> 
> http://www.markbaker.ca/2002/01/draft-baker-generic-xmlns-dispatch

If you buy the argument, (which I don't) that namespaces are the right
thing to trigger off of, then why wouldn't you just put the namespace in
the MIME type either using Simon's trick or using a URI->mime type
mapping: http://www.ietf.org/internet-drafts/draft-eastlake-cturi-03.txt
.

 Paul Prescod

Received on Monday, 4 February 2002 00:32:29 UTC