Eric van der Vlist's recent article on XML.com has drawn together several threads of thinking on the XML Infoset.

Van der Vlist's article augments a SAX stream with XML elements describing XML 1.0 constructs such as entity references. In response to the author's request for comments on XML-DEV, Jonathan Borden noted that many of the ideas were echoed in his XSet XML syntax model. Borden's technique allows an XML document to be fully expanded into its component parts, and then addressed with XPath/XPointer, right down to the level of "the 3rd character in an attribute value, or the 5th whitespace character in the ..."

Borden writes of his coming plans, which sound rather like a syntactic implementation of groves:

In many cases this [XML expansion] is too detailed. I've posted an early example of a way to "prune" this XSet tree into "meaningful" information ala the Infoset. ... What I'm currently developing are vocabularies to prune, condense and transform XSet plans into arbitrary XML models which can be addressed via XPath/XPointer.

Also in response to Van der Vlist, Richard Tobin mentioned an XML schema for the Infoset he had been working on, "with the intention of using it to compare the output of XML Schema implementations." The schema is available from Tobin's website.