[pro] best practices on type safety of/for generic functions

[Stas Boukarev]

John Morrison <john.nmi.morrison at gmail.com> writes:

> Hi All;
>
> This is probably a dumb question, but here goes.
>
> John Alan McDonald (Hi, John, if you're on this list!) has graciously
> consented to let me try to revive some almost 20 year old CL software (
> Arizona <http://home.comcast.net/~johnamcdonald/jamcdonald0/az93.pdf>).
> SBCL doesn't seem to like type-related declarations in defgeneric forms
> (per the spec<http://www.ai.mit.edu/projects/iiip/doc/CommonLISP/HyperSpec/Body/mac_defgeneric.html#defgeneric>).
> And there are a lot of defgenerics of the form:
>
> (declaim (declaration :returns)) ;; OK, only once, but then used
> repeatedly...
>
> (defgeneric interactor-role-cursor (interactor role)
>   (declare (type Interactor interactor)
>        (type (or Interactor-Role Null) role)
>        (:returns (type xlib:Cursor)))
>   (:documentation
>    "Returns the cursor to be used when this Role is current."))
>
> (defmethod interactor-role-cursor ((interactor Interactor)
>                    (role Interactor-Role))
>   "The default cursor"
>   (xlt:window-cursor (interactor-window interactor) :target))
>
> I did certainly benefit from some of the runtime errors generated as a
> result of the type declarations of some of the defmethods and defuns in my
> efforts to update the clx parts of the software.
>
> What is best practice, then, as regards trying to provide useful
> type-related information associated with generic functions?  Or have I been
> so brain-damaged by C++/Java/etc that I am thinking about the problem-space
> entirely the wrong way, and thus my solution-space question is entirely
> meaningless?
The portable equivalent would be
(declaim (ftype (function (interactor (or interactor-role null)) xlib:cursor)
                interactor-role-cursor))

-- 
With best regards, Stas.