Implicit conversion from float -9.899999999999991 to int loses precision