Avi Kelman wrote: > How can the name of an unused variable hurt performance so much? This changes the memory layout, which in turn may affect the heuristics, based on probabilistic algorithms. The real problem is that the code and data structures used are suboptimal. Mostly because it's way too branchy and the recursion is not deep enough to generate useful traces. Also, lots of inefficiencies, e.g. no use of local functions. And that rowminus stuff actually hurts performance. Basically, the JIT compiler gives up at a certain point, which depends on the heuristics. The timing you get depends on when and where that happens (see -jv). --Mike