Coupla quick comments - I accept most of your arguments for not making the language fully generalised (despite being a fan of both SPARQL & XPath). But what I'd caution though is making the language more specific than it needs to be, the other points being equal.
To avoid reinvention, a selective reuse of XPath constructs would seem to make sense, with added sugar for Atom. Can your proposal be systematically mapped to XPath? Having said that, although XML-orientation seems obvious given Atom (and the possibility of using URI path parts..? ok, I only skimmed your draft), I wouldn't entirely dismiss SPARQL-like features. Typed links *are* important in Atom, so some kind of triple patterns could be very useful. As another ref, I'd suggest checking out Microsoft's Project Astoria - it's essentially a RESTful front end on top of an entity-relation backend model. Like RDF but without URIs as entity/relation identifiers (baby...bathwater...hmm). But one thing I saw a demo of which looked very promising was their query URI construction. I can't find the spec proper, but there are some examples here: http://alexbarnett.net/blog/archive/2007/04/30/microsoft-codename-quot-astoria-quot-data-services-for-the-web.aspx Cheers, Danny. -- http://dannyayers.com
