bug#38206: 26.1; cl-prettyexpand incorrectly expands pushing to structures

[Lars Ingebrigtsen <larsi@gnus.org>]

Braun Gábor <braungb88@gmail.com> writes:

> (let* ((v test-slot)
>        (v (or (cl-block test-p
>                 (and (memq (type-of foo) cl-struct-test-tags) t))
>               (signal 'wrong-type-argument (list 'test foo))))
>        (v (aref foo 1)))
>   (\(setf\ cl-block\) (cons 1 (cl-block v v v)) v v v))
>
> The form in the output is obviously an incorrect expansion of
> (push 1 (test-slot foo)).

Yup.  However, if we give a FULL parameter to the function, it gives the
correct results:

(cl-prettyexpand '(push 1 (test-slot foo)) t)
=>
(progn
  (or (and (memq (type-of foo) cl-struct-test-tags) t)
      (signal 'wrong-type-argument (list 'test foo)))
  (let* ((v foo))
    (aset v 1 (cons 1 (aref v 1)))))

The definition is

(defun cl-prettyexpand (form &optional full)
  "Expand macros in FORM and insert the pretty-printed result.
Optional argument FULL non-nil means to expand all macros,
including `cl-block' and `cl-eval-when'."
[...]
    (setq form (macroexpand-all form
                                (and (not full) '((cl-block) (cl-eval-when)))))

The bug was introduced by a rewrite in 2012, I think, which changed the
implementation radically, which made it pass in that list as a totally
bogus ENVIRONMENT to macroexpand-all.

I think the right change here is to just deprecate the FULL parameter
and remove the 

                                (and (not full) '((cl-block) (cl-eval-when)))))

bit.  Stefan?

-- 
(domestic pets only, the antidote for overdose, milk.)
   bloggy blog: http://lars.ingebrigtsen.no