Stefan Monnier <monnier@iro.umontreal.ca> writes:

> > diff --git a/lisp/emacs-lisp/bytecomp.el b/lisp/emacs-lisp/bytecomp.el
> > index 966990bac9..418070fabe 100644
> > --- a/lisp/emacs-lisp/bytecomp.el
> > +++ b/lisp/emacs-lisp/bytecomp.el
> > @@ -3357,6 +3357,7 @@ byte-compile-check-variable
> >             (and od
> >                  (not (memq var byte-compile-not-obsolete-vars))
> >                  (not (memq var byte-compile-global-not-obsolete-vars))
> > +                (not (memq var byte-compile-lexical-variables))
> >                  (or (pcase (nth 1 od)
> >                        ('set (not (eq access-type 'reference)))
> >                        ('get (eq access-type 'reference))
>
> Yes!

Coming back to this now... Ok, then let's do that!

> > BTW, while I looked at this, I found this spurious lookup in
> > `byte-compile-lexical-variables':
> >
> > #+begin_src emacs-lisp
> > (defun byte-compile-form (form &optional for-effect)
> >   (let ((byte-compile--for-effect for-effect))
> >     (cond
> >      ((not (consp form))..10..)
> >      ((symbolp (car form))
> >       (let* ((fn (car form))..4..)
> >         (when (memq fn '(set symbol-value run-hooks..4..)
> >           (pcase (cdr form)
> >             (`(',var . ,_)
> >              (when (assq var byte-compile-lexical-variables) ;; <--- here
> >                (byte-compile-report-error
> >                 (format-message "%s cannot use lexical var `%s'" fn var)))
> >              ...)))))))))
> > #+end_src
> >
> > shouldn't the `assq' be `memq'?
>
> I think you're right.

Seems this has been treated meanwhile: Bug#44980 and commit ace6eba036
"Fix byte-compiler warning for failed uses of lexical vars", Stefan
Kangas.

> Both changes should ideally come with corresponding regression tests.

I tried to add an appropriate test for the remaining case, please check
the following patch (my first addition to the test suite) - is it what
you had in mind? - where I also try to give a less confusing (but
longer) compiler warning.  Does it look alright?