Implicit conversion from float -3.839999999999989 to int loses precision