Floating point tends to break all sorts of assumptions. For languages without static types, bignum integers and distinct operators for truncating vs FP division is probably best. This is how Python 3 works, plus Haskell and SML (with static types).