Andrew Haley writes: > > The paper describes the algorithm that the IBM jit uses. Sorry, this may not be the case. Forget I said that! Andrew.