Ok. Still, I'm under the impression that the Java glue generates machine code function stubs since JNI didn't consider the possibility that every called function doesn't have a unique entry point. I thought the same solution could be applied to Bills problem, although in a slightly different context.