Staring back at me today is something that Paul probably saw last year. But it's something that I found pretty cool because initially it looked like a mistake. Almost as cool as Tom starting to get SPARQL going.
Basically, there are three indices in JRDF: (0,1,2), (1,2,0) and (2,0,1). With the 0 equating to the subject, the 1 the predicate and 2 to the object.
Looking at a refactoring between these internal indices and triples (ordered by subject, predicate and object) there are these three mappings:
* (0,1,2) maps to (s,p,o) using (0,1,2),
* (1,2,0) maps to (s,p,o) using (2,0,1), and
* (2,0,1) maps to (s,p,o) using (1,2,0).
Using the second index as an example, it means that the internal representation maps the third element to the subject, the first element to predicate and the second element to the object.
The surprising thing I noticed, beside the other two indices not being tested correctly, was that the other indices ((2,1,0), (1,0,2), (0,2,1) - these are mentioned in Paul's post) don't seem to have this property. They all seem to map (s,p,o) like (0,1,2) - i.e. using themselves. Whereas, (1,2,0) mapped to itself gives (2,0,1) and (2,0,1) mapped to itself gives (1,2,0).