Raw

Leo Simons has posted about the design choices they've made - notably :

No bnodes

He gives an example, along with the advantage it brings to the code (less nested loops). It's a bit hard to extrapolate from though - the example involves a list, which is a fairly particular situation.

The idea of only using a subset of RDF/RDFS has been discussed a good few times before (orthogonally to the XML syntax permathread). Ian Davis brought it up in Crisis, which Shelley then tossed in oil. Basically reification seems more trouble than it's worth (especially when named graphs are available), containers are misleading, collections can be hard work. Containers/collections were a specific aspect of RDF the Metaweb folks mentioned as lacking. I think someone else has been playing with RDF without resources - or did I dream that...

All that stuff doesn't sound unreasonable to me, a lot of it could be seen as a side effect of preferring simpler, more performant designs. Having said that, I'm not entirely convinced by Leo's comparison in terms of making life easier for the coder - in other cases sugar may be an option, or a completely different approach, like Just Use SPARQL.

I would worry a bit about the side effects a policy of no bnodes in general might produce - either the domain models have to be bent to fit and/or logic being moved from the data model into the code, undermining the benefits of a declarative approach. (Whatever, I don't think any of it calls for any spec changes, the tools are generally mature enough that it is possible to pick & choose features).

[I wonder if there's an RDF and OWL 1.1 compatible profile (not Full) containing most of RDF/S, including existentials - but no reification or containers - plus FPs & IFPs, and that's about all. It would be good to have an 80/20 that could work alongside simple data models as well as the fanciest reasoners. Must read up.]

I also wonder if the recent revised abstract model for Dublin Core might help any with the example Ian used in his crisis post.

PS. Oops, nearly forgot the link there. Anyhow, forget the implementation details for now, go read this post from Bob DuCharme on Semantic data entry.