Re: [Emacs-diffs] master d895f6c: Fix Edebug specs for 'cl-defmethod' and 'defmethod'

Re: [Emacs-diffs] master d895f6c: Fix Edebug specs for 'cl-defmethod' and 'defmethod'

[Stefan Monnier]

> -             [ &optional keywordp ]     ; this is key :before etc
> -             list                       ; arguments
> +             [ &optional keywordp       ; this is key :before etc
> +               &optional stringp ]      ; :extra can be followed by a string

Actually, there can be any number of non-lists between the function name
and the arg list.  So it should probably be more like [ &rest atom ] or
something like that.


        Stefan

Re: [Emacs-diffs] master d895f6c: Fix Edebug specs for 'cl-defmethod' and 'defmethod'

[Dmitry Gutov]

Hi Stefan,

On 02.05.2017 17:08, Stefan Monnier wrote:
>> -             [ &optional keywordp ]     ; this is key :before etc
>> -             list                       ; arguments
>> +             [ &optional keywordp       ; this is key :before etc
>> +               &optional stringp ]      ; :extra can be followed by a string
> 
> Actually, there can be any number of non-lists between the function name
> and the arg list.  So it should probably be more like [ &rest atom ] or
> something like that.

Why not

-             [ &optional keywordp       ; this is key :before etc
-               &optional stringp ]      ; :extra can be followed by a 
string
+             [ &rest              ; multiple qualifiers are allowed
+               &or keywordp       ; this is key :before etc
+               stringp ]          ; :extra can be followed by a string

?

Do we anticipate any non-keyword, non-string qualifiers?

Re: [Emacs-diffs] master d895f6c: Fix Edebug specs for 'cl-defmethod' and 'defmethod'

[Stefan Monnier]

> Do we anticipate any non-keyword, non-string qualifiers?

The CLOS spec allows any non-list value and cl-generic.el implements the
same rule.  Whether some code somewhere will ever make use of that extra
flexibility, I don't know.  I can imagine numbers and non-keyword
symbols being used, and since cons cells aren't allowed, I could also
imagine vectors being handy for complex cases.  I'm hard pressed to
imagine use of anything beyond that, OTOH.


        Stefan

Re: [Emacs-diffs] master d895f6c: Fix Edebug specs for 'cl-defmethod' and 'defmethod'

[Dmitry Gutov]

On 04.05.2017 18:50, Stefan Monnier wrote:

> The CLOS spec allows any non-list value and cl-generic.el implements the
> same rule.  Whether some code somewhere will ever make use of that extra
> flexibility, I don't know.  I can imagine numbers and non-keyword
> symbols being used, and since cons cells aren't allowed, I could also
> imagine vectors being handy for complex cases.  I'm hard pressed to
> imagine use of anything beyond that, OTOH.

I see, thanks. Applied your original suggestion.

Re: [Emacs-diffs] master d895f6c: Fix Edebug specs for 'cl-defmethod' and 'defmethod'

[Dmitry Gutov]

On 02.05.2017 17:08, Stefan Monnier wrote:
>> -             [ &optional keywordp ]     ; this is key :before etc
>> -             list                       ; arguments
>> +             [ &optional keywordp       ; this is key :before etc
>> +               &optional stringp ]      ; :extra can be followed by a string
> 
> Actually, there can be any number of non-lists between the function name
> and the arg list.  So it should probably be more like [ &rest atom ] or
> something like that.

By the way, do we want to support methods with zero arguments here? nil 
is an atom, and not a list.

Re: [Emacs-diffs] master d895f6c: Fix Edebug specs for 'cl-defmethod' and 'defmethod'

[Stefan Monnier]

>>> -             [ &optional keywordp ]     ; this is key :before etc
>>> -             list                       ; arguments
>>> +             [ &optional keywordp       ; this is key :before etc
>>> +               &optional stringp ]      ; :extra can be followed by a string
>> 
>> Actually, there can be any number of non-lists between the function name
>> and the arg list.  So it should probably be more like [ &rest atom ] or
>> something like that.

> By the way, do we want to support methods with zero arguments here? nil is
> an atom, and not a list.

IIRC CLOS does not allow 0-arg methods (and it allows nil as a qualifier),
whereas cl-defmethod uses `listp` to distinguish args from qualifiers,
so it doesn't allow nil as qualifier (tho I suspect it doesn't handle
a 0-arg method "correctly" either).


        Stefan

Re: [Emacs-diffs] master d895f6c: Fix Edebug specs for 'cl-defmethod' and 'defmethod'

[Dmitry Gutov]

On 10.05.2017 18:18, Stefan Monnier wrote:

> IIRC CLOS does not allow 0-arg methods (and it allows nil as a qualifier),
> whereas cl-defmethod uses `listp` to distinguish args from qualifiers,
> so it doesn't allow nil as qualifier (tho I suspect it doesn't handle
> a 0-arg method "correctly" either).

Seems to handle it okay, these both work:

(cl-defgeneric foo ()
   (1+ 2))

(cl-defmethod foo ()
   (1+ 4))

The latter redefines the default method, but that's to be expected.

If we want to support that kind of cl-defmethod in edebug, apparently we 
shold replace [&rest atom] with [&rest [&not listp]].

Re: [Emacs-diffs] master d895f6c: Fix Edebug specs for 'cl-defmethod' and 'defmethod'

[Dmitry Gutov]

On 11.05.2017 0:49, Dmitry Gutov wrote:

> If we want to support that kind of cl-defmethod in edebug, apparently we 
> shold replace [&rest atom] with [&rest [&not listp]].

Or maybe [&rest &not null atom].