My first point was the most critical, so let me just add a couple of comments/questions.

First, on identifying (and therefore being able describe relationships to) non-document objects, given that it seems your concerns about RDFa were largely about syntax, is there not a better solution to this requirement than an about property? I’m imagining that this particular solution will force some pretty awkward syntax.

Second, I’m unclear from the examples, at least, but how does one establish a URI predicate on a relation? E.g. I want to say that “Jane Doe” co-authored a paper with me, where Jane has a profile page she uses to identify her? So I want an a element that links to that, and I want to embed a property that says she’s an author of that document. How can I do that with this proposal?

Reading over Ian’s list email again, I think I have the next assumption wrong, and that one encodes the predicate using a property attribute on the “a” element. Am not totally sure though.

From what I gather, the current answer is I can’t (since rel values are only string tokens; not URIs). Under the section on converting to RDF, item 2.9 just seems bad, since it severely constrains what you can say about the data. For sake of argument, is it possible to allow the rel/rev values to be URIs?

The prefix stuff is less important to me, but my point is that the reverse DNS identifiers seems an arbitrary inclusion that doesn’t seem warranted by the use cases.

Also, it seems there’s been work in the RDFa community on removing the requirement for prefixes by using things like profiles (see also this similar work on a JSON serialization of RDF; basic approach is you define the mapping of URI to string token somewhere apart from the actual encoded properties). Do you see the microdata proposal being able to accommodate that potentially productive evolution?

Finally, on “item” vs. “typeOf” I guess this is exactly where I think you’re confusing type and identity. You say in that email you link to that What we need is a way to say that a property is really a new set of name-value pairs. Yes, correct. But in RDFa, this is exactly what the “about” attribute does; it switches the context of the statements. It’s just that in your proposal and associated examples, you are always having the impulse to want to create blank nodes; e.g. to not give the object you’re describing a global ID.

So to boil down my point here, you are looking to the wrong mechanism to solve the problem you identify in the quote above. Suffice it to say I think this area needs work.

I’m not sure how it follows that Java-like identifiers (com.example.foo) mean we should allow prefixes. Prefixes in general have proved quite confusing to most authors — the problem with separating where the prefix is declared and where the prefix is used results in people breaking pages during copy-and-paste, getting the declarations wrong, etc. Even implementors seem to have a hard time with them, with implementors hard-coding individual prefixes, using regexps to detect particular keywords, etc.

item=”" is similar to typeOf; the reason I didn’t use just typeOf is that it looks weird to say “typeOf” on its own. I discuss this in more detail in the e-mail: http://lists.whatwg.org/htdig.cgi/whatwg-whatwg.org/2009-May/019681.html

HTH!