[Openmcl-devel] Warning about unused arguments in generic function methods

[Hans Hübner]

Currently, you get a warning only for parameters that are implicitly
specialized to the T class.  If you have an explicit specializer, no
warning is generated:

CL-USER> (defgeneric foo (a b))
#<STANDARD-GENERIC-FUNCTION FOO #x300045AFEBDF>
CL-USER> (defmethod foo ((a t) (b t)) t)
#<STANDARD-METHOD FOO (T T)>
CL-USER> (defmethod foo (a b) t)

;Compiler warnings :
;   Unused lexical variable A, in (FOO (T T)) inside an anonymous lambda form.
;   Unused lexical variable B, in (FOO (T T)) inside an anonymous lambda form.

Following your suggestion, the following patch may do:

Index: macros.lisp
===================================================================
RCS file: /usr/local/tmpcvs/openmcl-dev/ccl/lib/macros.lisp,v
retrieving revision 1.118
diff -c -r1.118 macros.lisp
*** macros.lisp	15 Oct 2007 04:10:54 -0000	1.118
--- macros.lisp	14 Nov 2007 12:29:20 -0000
***************
*** 1679,1685 ****
            (when inner-list
              (push `(declare ,@(nreverse inner-list)) inner)))))
      (values (nreverse outer) (nreverse inner))))
! 		

  (defun parse-defmethod (name args env)
    (validate-function-name name)
--- 1679,1686 ----
            (when inner-list
              (push `(declare ,@(nreverse inner-list)) inner)))))
      (values (nreverse outer) (nreverse inner))))
!
! (defvar *DEFMETHOD-DECLARE-ALL-SPECIALIZED-ARGUMENTS-IGNORABLE* t)

  (defun parse-defmethod (name args env)
    (validate-function-name name)
***************
*** 1708,1713 ****
--- 1709,1716 ----
                          (push `(type ,p ,(%car parameters)) types)))
                       (t (signal-program-error "Illegal arg ~S" p))))
                (t
+                (when *defmethod-declare-all-specialized-arguments-ignorable*
+                  (push p refs))
                 (push p parameters)
                 (push t specializers-form)
                 (push t specializers)))))


On Nov 14, 2007 10:27 AM, Gary Byers <gb at clozure.com> wrote:
> For a parameter specialized (implicitly or explicitly) to the
> T class, it might be the case that that parameter is strictly
> used for discrimination, but it's at least equally likely that
> it's "just an argument".  For example:
>
> (defmethod stream-write-char ((s telnet-stream) char)
>    s)
>
> I think that I would want a warning telling me that I'd forgotten
> to reference CHAR; if I hadn't referenced S either ... well, it's
> at least somewhat plausible that the specialized parameter was
> used for discrimination only (though I think that I'd like to
> know if I wrote a STREAM-WRITE-CHAR method that didn't reference
> either of its arguments.)
>
> [If I actually -wanted- to ignore the character - as in
>
> (defmethod stream-write-char ((s null-stream) char)
>    (declare (ignore char)))
>
> the IGNORE declaration doesn't seem too painful and seems to make the
> intentional ignoring of the argument clear]
>
> It's certainly possible to think of examples where the failure to
> reference a parameter that's specialized to the T class is in no way
> anomalous (likely because such parameters were used solely for
> discrimination.)  I can't think of a way for the compiler to guess
> that when compiling a method body (or at least no way for it to guess
> right consistently): that depends on contracts and conventions and (to
> some extent at least) on coding style.
>
> It's not clear that method qualifiers change this significantly,
> but I do wonder about cases like:
>
> (defmethod stream-write-char ((s some-stream-class) c)
>    (call-next-method))
>
> since the 0-arg CALL-NEXT-METHOD arguably references the explicit
> arguments.
>
> DEFMETHOD could macroexpand into something which declares all
> required method arguments to be IGNORABLE (it currently does
> that only for those things that're specialized to something
> other than the T class), perhaps under the control of some
> global variable (*DEFMETHOD-DECLARE-ALL-SPECIALIZED-ARGUMENTS-IGNORABLE*
> or some such.)  I don't know how you'd know how to set that so
> that you got warnings in cases like the STREAM-WRITE-CHAR but
> not in the cases that you find annoying.
>
>
>
> On Wed, 14 Nov 2007, Hans Hübner wrote:
>
> > Hi,
> >
> > is there a way to make OpenMCL stop complaining about unused arguments
> > in generic function methods?  I am fine with the warning in standard
> > function definitions, but with methods it makes less sense because
> > often, the arguments are present just to match the generic function
> > signature and rightfully ignored.  This is especially true with
> > :before and :after methods, where I find it very annoying to have to
> > explicitly ignore unused arguments
> >
> > -Hans
> > _______________________________________________
> > Openmcl-devel mailing list
> > Openmcl-devel at clozure.com
> > http://clozure.com/mailman/listinfo/openmcl-devel
> >
> >