On a theorem of the alternative and saddle-point conditions in nonsmooth optimization