dispatching on keywords

[Mark Nahabedian <naha@RIVERSIDE.SCRC.Symbolics.COM>]

    Date: Fri, 13 Nov 1992 11:45 EST
    From: davis@ilog.ilog.fr (Harley Davis)

    I think the idea of lexicographical ordering of class names for
    computing method specificity is silly --- especially in a system where
    class names are optional and documentary.

    However, I think there is something which can be saved from the idea
    of dispatching on keywords and optional arguments.  I think it should
    be allowed only if the default value is specified in the generic
    function definition rather than in the method definition.  This would
    not create the sort of ambiguities which Moon and Fahlman mentioned.

    My proposal would be that if a default value is specified in a generic
    function, it cannot also be specified in one of the gf's methods.  If
    any method specifies a default value, the generic function cannot.

I think you also need the restriction that if any method specifies a
default for an argument then no method can specialize that argument.
You can't change the arguments after the applicable methods have been
determined.

    Example:

    (defgeneric foo (x #optional (y 1))

    (defmethod foo (x #optional (y <integer>)) 
      ;; default method
      )

    (defmethod foo (x #optional (y <string>)) 
      ;; could be called if y were supplied by the caller
      )

    Obviously there are major syntactic problems to be solved.

... because the syntax for specializing required arguments conflicts
with the syntax for defaulting optional arguments.

I don't understand the motivation for this discussion.  I have not yet
needed such a feature.  I have always been able to employ a trivial
workaround on the rare occasions when what you're arguing for ha scome
up.  (Sorry to be using Lisp syntax here, I don't have a Dylan spec
handy).

(defun foo (x &optional (y integer))
  (foo-1 x y))

(defgeneric foo-1 (x y))

(defmethod foo-1 (x (y integer))
  ...)

(defmethod foo-1 (x (y string))
  ...)