On an argument of J.--F. Cardoso dealing with perturbations of joint diagonalizers