Comments on trace anomaly matching