When an object of integral type is converted to a floating type and the original value cannot be represented exactly, the result is either the next higher or the next lower representable value.