> And how about calling alloca in a varargs function that has an eliminated > frame pointer and incoming args space? That could be a good way to expose > latent bugs...... <g> > The backends usually say the frame_pointer is needed if alloca is used. Graham