Implicit conversion from float -3.799999999999997 to int loses precision