This does not make sense: floating point numbers are not really "approximate". For example, small integer arithmetic using floats is guaranteed to be exact. Your language would be very confusing for this common use case.