why precision is essential in floatdouble comparison