Accidental neat trick

[kpreid]

In four easy steps!

1. Decide to write a toy RDF-processing library in E.

2. Mangle Improve the RDF data model: instead of having the three disjoint types URI, blank node, and literal, everything in a graph is like a blank node and has 0 or more names which may be URIs, literals, or any E object—which will, you imagine, be used mainly for representing datatyped literals.

3. Decide to have the basic graph class not know about any types of names, and allow any object, with only conventions for URIs and literals.

4. Realize that a trivial naive smusher can be written using this data model: for each triple whose predicate is an owl:InverseFunctionalProperty, add a name for the subject of the triple consisting of the predicate and object nodes, and any duplication must then result in the nodes being unified. This also preserves multiple URIs for the same subject without requiring duplicate triples.

Of course, I have no idea yet whether this model and library will be good for anything useful yet. (If nothing else, it will certainly be horribly slow.)

I should collect some use cases.