bug#25929: 25.2; map-delete doesn't delete permanently 1st alist elt

[Lars Ingebrigtsen <larsi@gnus.org>]

Michael Heerdegen <michael_heerdegen@web.de> writes:

> Stefan Monnier <monnier@iro.umontreal.ca> writes:
>
>> >     (macroexp-let2 nil key
>> >       `(progn
>> >          (map--dispatch ,map
>> >            :list (setf (alist-get ,key ,map nil t) nil)
>> >            :hash-table (remhash ,key ,map)
>> >            :array (and (>= ,key 0)
>> >                        (<= ,key (seq-length ,map))
>> >                        (aset ,map ,key nil)))
>> >          ,map)))
>>
>> Note that this will make it pretty much impossible to use
>> cl-generic dispatch.
>
> It will also break cases where MAP is an expression that is not a
> symbol:
>
> #+begin_src emacs-lisp
> (let ((thing (cons 'tag (list (cons 0 1) (cons 2 3)))))
>   (map-delete (cdr thing) 0))
> ==> nil
>
> (map-delete (list (cons 0 1) (cons 2 3)) 0)
> ==> if: Symbol's function definition is void: \(setf\ list\)
> #+end_src

If MAP isn't a symbol, then map-delete could avoid doing the setf,
couldn't it?

I think that should make for a pretty useful form.  You can both say

...
(map-delete foo 'bar)
...

and

(setq foo (cons (map-delete (get-a-list) 'bar) 'zot))

and not be surprised at the results.  It's certainly more fun than the
current (setq foo (delete 'bar foo)) we have all over the place, even if
it's something of a departure from how functions like this has
traditionally worked in various Lisps.

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