bug#66908: Exposing more public nadvice API

[Stefan Monnier]

> Sure, in Philip's package do-at-point, a function defined to "act" on
> the `thing' at point are given different arguments depending on the
> minimum number of arguments required by the function:

Ah :-(

So a kind of "unimplementable semantics" for DWIM purposes.

>         (pcase (car (func-arity func))
>                     ^^^^^^^^^^^^^^^^^
>           (0 (funcall func))
>           (1 (funcall func (buffer-substring (car bound) (cdr bound))))
>           (2 (funcall func (car bound) (cdr bound)))
>           (_ (error "Unsupported signature: %S" func)))))

I recommend:

    (condition-case nil
        (funcall func (car bound) (cdr bound))
      (wrong-number-of-arguments
       (condition-case nil
           (funcall func (buffer-substring (car bound) (cdr bound)))
         (wrong-number-of-arguments
          (funcall func)))))

:-)

Works with advice and other wrappers without having to worry about
`indirect-function`, autoloading, etc...

>>>     (func-arity (advice--cd*r (indirect-function 'xref-find-definitions))) 
>>> ;; ⇒ (1 . 1)
>>> which is the right return value.  It might be nice to not have to call
>>> `indirect-function' here for the "global" function but you can be a
>>> better judge of that.
>> Don't know what you mean by "global" function.
> By "global", I mean the new global function advice-cd*r or somesuch that
> might eventually be added from this discussion.

Ah, I see.  I don't think `advice-cd*r` should follow aliases in
general.  But indeed, you may have an `advice` object whose
`advice-cd*r` is a symbol, whose definition is another advice object,
etc...

Another good reason to prefer the `condition-case` approach :-)

>> By "autoloaded" do you mean "setup to be loaded on demand but not yet
>> loaded", or do you mean "had been setup to be loaded on demand and has
>> been loaded already"?
> The former obviously.

In that case, `indirect-function` would not see the advice, then
`func-arity` would cause the target to be (auto)loaded, during which
a previously pending advice could be installed and it would return the
dreaded (0 . many) from the advice it sees in the definition.

Again, using `condition-case` side steps the issue.


        Stefan