A remark on robustness and weak continuity of M-estimators