Michael,
> I wouldn't normally recommend using an XML database unless it is natural to think about the data as being a set of documents.
I'm increasingly coming to that position. It's part of the reason I've been quietly slipping into the RDF world the last few years: a significant portion of the data that I work with tends to be heavily linked resource descriptions, though often with a documentish context to them. These structures tend to be partially normalized, making them awkward to work with in a standard SQL database, and because they are enterprise-centric, the necessity for managing a global identity space is fairly critical, so RDF seems a natural fit. Having said that, RDF does not handle narrative structures well, because order is no longer an implicit concept.
I think the next stage is the emergence of hybrid data systems, ones where you have some degree of control over the normalization process, but can nonetheless make use of n-tuples while, when necessary, managing narrative integrity. Daniela Florescu's vision with jsoniq goes in the right direction - abstract those portions of XQuery that's bound to the processing model away from XML exclusively, and treat the structures as, well, abstract data structures. The XDM and JSDM are both reasonably well understood at a data algebra level, RDF is a generalized graph that can present itself in any modality but sits on an ntuple index as opposed to the 2-tuple indices typical of most NoSQL architectures.
Of course, I could be wrong - it seems like a logical progression, so no doubt some large company with a vested interest will do their hardest to make sure it doesn't happen. Still, I think the approach is sound.
