Knuth's Algorithm D, plus the annoying question: when is the trial quotient correction actually wrong?