Mark Dickinson: > This is probably the only sane way to deal with differences in > platform behaviour when doing float divisions. What Python run on a CPU that doesn't handle the nan correctly? Bye, bearophile