Comments on the non-Abelian Stokes theorem