Implicit conversion from float -23.999999999999986 to int loses precision