Even decimal floating point is a bad idea (for dealing with money) since you still can't represent a subset of rational numbers without approximation and without introducing rounding error during some calculations. It's just a different subset than what binary floating point can represent without approximation.