Implicit conversion from float -34.25 to int loses precision