Implicit conversion from float -17.125 to int loses precision