Re-evaluating Method-Level Bug Prediction