We have some tests already, but we are lacking a benchmark suite to test different algorithms and implementations against. We should look into benchmarks, timing, and making sure the optimizer does The Right Thing (tm) there.