Paul Moore wrote: > But no-one manually raises NameError, so Ken's example wouldn't work > with "real" NameErrors. Part of his suggestion was that core and stdlib code would move towards using attributes. The NameError example makes sense in that context. -- Greg