OK, I have a patch that does much better for the case where all variables are bound. Unfortunately, it doesn't do anything special if there are unbound values. In that case it just reverts to the linear search again.
It'd be possible to build indexes based on a knowledge of what *could* be unbound, but that was going to take more static analysis than I had time for last night. :-) Anyway, it seems to be running correctly, but I want to run a decent set of tests against it before posting the patch. It seems that there aren't any tests for MINUS in the SPARQL test suite, so it looks like I need to do that too. :-) Paul On Wed, Jun 13, 2012 at 5:02 PM, Andy Seaborne <[email protected]> wrote: > On 13/06/12 21:52, Alex Hall wrote: >> >> The difficult part there is going to be finding the common >> variables. > > > (haven't had time to catch up and all the interesting email about this on > both the lists yet ...) > > Just on this point - much of that can be done by static analysis. It > provides a superset of the variables as some may not actually be bound for a > given row. > > Andy >
