bug#33170: 27.0.50; interactive spec with cl-defgeneric/method

bug#33170: 27.0.50; interactive spec with cl-defgeneric/method

[Alex Branham]

[-- Attachment #1: Type: text/plain, Size: 476 bytes --]

I'm struggling to understand how to make functions defined with
cl-defgeneric interactive:

(require 'cl-lib)

(cl-defgeneric my/test ()
  (interactive)
  (message "generic"))

If you evaluate that you can do M-x my/test, as expected. However, as
soon as you evaluate this:

(cl-defgeneric my/test (&context (major-mode emacs-lisp-mode))
  (interactive)
  (message "method"))

M-x my/test no longer works.

Is this a bug or am I misunderstanding how this works?

Thanks,
Alex

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 487 bytes --]

bug#33170: 27.0.50; interactive spec with cl-defgeneric/method

[Eli Zaretskii]

> From: Alex Branham <alex.branham@gmail.com>
> Date: Fri, 26 Oct 2018 17:19:44 -0500
> 
> I'm struggling to understand how to make functions defined with
> cl-defgeneric interactive:
> 
> (require 'cl-lib)
> 
> (cl-defgeneric my/test ()
>   (interactive)
>   (message "generic"))
> 
> If you evaluate that you can do M-x my/test, as expected. However, as
> soon as you evaluate this:
> 
> (cl-defgeneric my/test (&context (major-mode emacs-lisp-mode))
>   (interactive)
>   (message "method"))
> 
> M-x my/test no longer works.

'&context' is documented only as part of cl-defmethod, so why are you
trying to use it with cl-defgeneric?

bug#33170: 27.0.50; interactive spec with cl-defgeneric/method

[Alex Branham]

[-- Attachment #1: Type: text/plain, Size: 1086 bytes --]


On Sat 27 Oct 2018 at 01:54, Eli Zaretskii <eliz@gnu.org> wrote:

>> From: Alex Branham <alex.branham@gmail.com>
>> Date: Fri, 26 Oct 2018 17:19:44 -0500
>> 
>> I'm struggling to understand how to make functions defined with
>> cl-defgeneric interactive:
>> 
>> (require 'cl-lib)
>> 
>> (cl-defgeneric my/test ()
>>   (interactive)
>>   (message "generic"))
>> 
>> If you evaluate that you can do M-x my/test, as expected. However, as
>> soon as you evaluate this:
>> 
>> (cl-defgeneric my/test (&context (major-mode emacs-lisp-mode))
>>   (interactive)
>>   (message "method"))
>> 
>> M-x my/test no longer works.
>
> '&context' is documented only as part of cl-defmethod, so why are you
> trying to use it with cl-defgeneric?

Sorry, that was a typo. The same thing happens if you use defmethod:

(require 'cl-lib)

(cl-defgeneric my/test ()
  (interactive)
  (message "generic"))

here, M-x my/test works

(cl-defmethod my/test (&context (major-mode emacs-lisp-mode))
  (interactive)
  (message "method"))

Here, M-x my/test does not work.

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 487 bytes --]

bug#33170: 27.0.50; interactive spec with cl-defgeneric/method

[Eli Zaretskii]

> From: Alex Branham <alex.branham@gmail.com>
> Cc: 33170@debbugs.gnu.org
> Date: Sat, 27 Oct 2018 07:40:11 -0500
> 
> (require 'cl-lib)
> 
> (cl-defgeneric my/test ()
>   (interactive)
>   (message "generic"))
> 
> here, M-x my/test works
> 
> (cl-defmethod my/test (&context (major-mode emacs-lisp-mode))
>   (interactive)
>   (message "method"))
> 
> Here, M-x my/test does not work.

But "M-: (my/test) RET" does.

I don't think generic functions can be interactive.  At least it is
not documented anywhere that they can be.

bug#33170: 27.0.50; interactive spec with cl-defgeneric/method

[Alex Branham]

On Sat 27 Oct 2018 at 09:47, Eli Zaretskii <eliz@gnu.org> wrote:

>> (require 'cl-lib)
>>
>> (cl-defgeneric my/test ()
>>   (interactive)
>>   (message "generic"))
>>
>> here, M-x my/test works
>>
>> (cl-defmethod my/test (&context (major-mode emacs-lisp-mode))
>>   (interactive)
>>   (message "method"))
>>
>> Here, M-x my/test does not work.
>
> But "M-: (my/test) RET" does.

Right. I'd like them to be interactive though so I can e.g. use them
from a keymap.

> I don't think generic functions can be interactive.  At least it is
> not documented anywhere that they can be.

Ah, that's unfortunate. Perhaps this limitation should be documented
somewhere?

Thanks,
Alex

bug#33170: 27.0.50; interactive spec with cl-defgeneric/method

[Eli Zaretskii]

> From: Alex Branham <alex.branham@gmail.com>
> Cc: 33170@debbugs.gnu.org
> Date: Sat, 27 Oct 2018 09:50:26 -0500
> 
> >> Here, M-x my/test does not work.
> >
> > But "M-: (my/test) RET" does.
> 
> Right. I'd like them to be interactive though so I can e.g. use them
> from a keymap.

You could have a single command that invoked my/test, I think that
would do what you want.

> > I don't think generic functions can be interactive.  At least it is
> > not documented anywhere that they can be.
> 
> Ah, that's unfortunate. Perhaps this limitation should be documented
> somewhere?

If Stefan (CC'ed) says it's a limitation and not a bug, I will
document it.

bug#33170: 27.0.50; interactive spec with cl-defgeneric/method

[Alex Branham]

[-- Attachment #1: Type: text/plain, Size: 208 bytes --]


On Sat 27 Oct 2018 at 11:11, Eli Zaretskii <eliz@gnu.org> wrote:

> You could have a single command that invoked my/test, I think that
> would do what you want.

Ah, good idea. That does work, thanks!

Alex

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 487 bytes --]

bug#33170: 27.0.50; interactive spec with cl-defgeneric/method

[Stefan Monnier]

>> > I don't think generic functions can be interactive.  At least it is
>> > not documented anywhere that they can be.
>> Ah, that's unfortunate. Perhaps this limitation should be documented
>> somewhere?
> If Stefan (CC'ed) says it's a limitation and not a bug, I will
> document it.

AFAIC it's a limitation.  We could support interactive specs, but I'm
really not convinced it's worth the trouble.  If using 2 functions (one
generic and one interactive, which calls the generic one) is really
annoying for some reason, I guess you could use

    (put '<GENERIC> 'interactive-form <FORM>)

but I think this property should be deprecated, so don't tell anyone
(especially don't tell me) that I even mentioned it to you,


        Stefan

bug#33170: 27.0.50; interactive spec with cl-defgeneric/method

[Alex Branham]

[-- Attachment #1: Type: text/plain, Size: 1138 bytes --]


On Sat 27 Oct 2018 at 16:49, Stefan Monnier <monnier@IRO.UMontreal.CA> wrote:

> AFAIC it's a limitation.  We could support interactive specs, but I'm
> really not convinced it's worth the trouble.  If using 2 functions (one
> generic and one interactive, which calls the generic one) is really
> annoying for some reason, I guess you could use
>
>     (put '<GENERIC> 'interactive-form <FORM>)
>
> but I think this property should be deprecated, so don't tell anyone
> (especially don't tell me) that I even mentioned it to you,

Thanks for the response. Seems like the two-function solution is the way
to go I guess. I do think this should be documented as a limitation
somewhere though. Perhaps a note at the end of that section in the
manual along the lines:

Note that a current limitation of generic functions is that they may not
be interactive. The recommended way of working around this is to define
an interactive function that simply calls the generic function.

That's not strictly true of course since (cl-defgeneric foo ()
(interactive) (message "foo")) works just fine. It's adding methods that
seems to mess it up.

Alex

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 487 bytes --]

bug#33170: 27.0.50; interactive spec with cl-defgeneric/method

[Eli Zaretskii]

> From: Alex Branham <alex.branham@gmail.com>
> Cc: Eli Zaretskii <eliz@gnu.org>, 33170@debbugs.gnu.org
> Date: Mon, 29 Oct 2018 09:18:53 -0500
> 
> On Sat 27 Oct 2018 at 16:49, Stefan Monnier <monnier@IRO.UMontreal.CA> wrote:
> 
> > AFAIC it's a limitation.  We could support interactive specs, but I'm
> > really not convinced it's worth the trouble.  If using 2 functions (one
> > generic and one interactive, which calls the generic one) is really
> > annoying for some reason, I guess you could use
> >
> >     (put '<GENERIC> 'interactive-form <FORM>)
> >
> > but I think this property should be deprecated, so don't tell anyone
> > (especially don't tell me) that I even mentioned it to you,
> 
> Thanks for the response. Seems like the two-function solution is the way
> to go I guess. I do think this should be documented as a limitation
> somewhere though.

Done.

bug#33170: 27.0.50; interactive spec with cl-defgeneric/method

[Stefan Monnier]

>> Thanks for the response. Seems like the two-function solution is the way
>> to go I guess. I do think this should be documented as a limitation
>> somewhere though.
> Done.

Should we add a warning for it?


        Stefan


diff --git a/lisp/emacs-lisp/cl-generic.el b/lisp/emacs-lisp/cl-generic.el
index fb4f3ab680..025653de70 100644
--- a/lisp/emacs-lisp/cl-generic.el
+++ b/lisp/emacs-lisp/cl-generic.el
@@ -355,6 +355,9 @@ cl-generic-define-context-rewriter
                                    . ,(lambda () spec-args))
                                  macroexpand-all-environment)))
       (require 'cl-lib)        ;Needed to expand `cl-flet' and `cl-function'.
+      (when (interactive-form (cadr fun))
+        (message "Interactive forms unsupported in generic functions: %S"
+                 (interactive-form (cadr fun))))
       ;; First macroexpand away the cl-function stuff (e.g. &key and
       ;; destructuring args, `declare' and whatnot).
       (pcase (macroexpand fun macroenv)

bug#33170: 27.0.50; interactive spec with cl-defgeneric/method

[Eli Zaretskii]

> From: Stefan Monnier <monnier@IRO.UMontreal.CA>
> Cc: Alex Branham <alex.branham@gmail.com>, 33170-done@debbugs.gnu.org
> Date: Tue, 30 Oct 2018 08:48:49 -0400
> 
> >> Thanks for the response. Seems like the two-function solution is the way
> >> to go I guess. I do think this should be documented as a limitation
> >> somewhere though.
> > Done.
> 
> Should we add a warning for it?

No objections from me.

bug#33170: 27.0.50; interactive spec with cl-defgeneric/method

[Stefan Monnier]

>> Should we add a warning for it?
> No objections from me.

Done in `master`.


        Stefan