Klotz, Leigh wrote:
Bill, Thank you for the answer.
I'm being cautious here, because vocabulary integration is one of my main concerns in the direction Atom is taking, and I hate to see everything hard-coded with special deference to particular HTML tags. If we can't solve the problems without recourse to spec changes, we won't be building an extensible standard.
Could you let me know why RNG can't be normative? Is it written down somewhere in the RFC?
It is written down. There is consensus in the WG that the spec text be the final word.
There's normative BNF in various IETF RFCs and I don't see much difference, especially since RNC looks like BNF and it's now an ISO standard.
I do tend to disagree on the terminology issue, still, though.
I think that saying we have an "XML-in-XML" problem leads people to think of encapsulation and semantic-blind content carrying (think base64 or CDATA), whereas calling it "mixed vocabulary" implies solutions that are at the semantic level, and I believe lead to extensibility.
Extensibility so far in Atom has been firstly along the axis of particular attributes/values that applies to the format itself. Second is along the axis of adding names from other namespaces as children of the entry element making Atom a general purpose metadata container (the entry then is much like a dictionary). But while XHTML and text is an issue atm, it's worth pointing out that atom:content can carry arbitrary XML.
[After some thought, I'll stand by the XML in XML thing while believing it open up a level of debate that is maybe not helpful right now.]
Dave Orchard and Dare Obasanjo have written eloquently on designing languages for for integration and extensibility [1] [2] [3]. The articles point out some obstacles in using XML Schema so I suggested RNG; although prose would work, it's more, well, wordy, and open to interpretation in the implementation.
They have.The problem for us going that route is as I see it that we'll risk imposing processing constraints and coordination on producers and consumers that will be rejected in practice.
[My personal view is that this speaks to modularity and not extensibility; for example things like RDF, or say Lisp, have a very notion of what an extension is. Being able to mix and match lexically and between versions is not what I think about first when it comes to extension, but I've been told that's not a common interpretation.]
cheers Bill
