No portable way to detect string-to-float conversion errors in standard C