Re: Problem with let/cl-letf binding stuff with org-capture

[Arthur Miller <arthur.miller@live.com>]

Bruno Barbier <brubar.cs@gmail.com> writes:

> Arthur Miller <arthur.miller@live.com> writes:
>
>> Bruno Barbier <brubar.cs@gmail.com> writes:
>>
>>> Arthur Miller <arthur.miller@live.com> writes:
>>>
>>> The hook `org-capture-mode-hook' will be run in your special
>>> capture buffer. You can override the "C-c C-c" binding only there.
>>
>> Yes and in every other capture buffer
>
> No. If you modify the hook only during your call to 'org-capture', it
> will be called only once in your buffer.
>
>>> Even if I could let bind the function at the right time, I would avoid
>>> that solution, as I can't garantuee that this global hack will not break
>>> other parts of Emacs (other captures, output filters, threads, timers,
>>> etc.).
>>
>> Why do you think it will break other parts? This is not a global hack, on contrary it
>> exactly tries to prevent to be "global" in entire Emacs, by let-binding a name
>> to a local lambda, which becomes "global" only in that buffer. If that
>> explains.
>
> You are assigning a local value to the global binding. Everything in
> Emacs will use your functions until you exit the cl-letf. It's like if
> you were using 'fset'.

Yes, you are totally correct; unfortunately binding with cl-flet or cl-labels,
does not work, so binding the global is the only one that works.

> Here is an example that uses cl-letf. Note that the call to
> async-shell-command is outside the "local" binding, yet, the cl-letf
> breaks it. You should try this in an other Emacs, just in case.
>
> (defun oops ()
>   (let ((v (async-shell-command "date" "!sh async")))
>     (cl-letf
>         (((symbol-function 'comint-output-filter)
>           (lambda (proc string)
>             (message "async-shell-command is using my binding: %s" string)
>             (read-string "What's the password?"))))
>       (read-string "what: ")
>       )))
> (oops)

Yes, I definitely agree with you about the concerns, which are certainly
valid, but you would get same effect with advice, it is not different at
all. The difference is that let-binding is automatically removed and installed,
and only active during the cl-letf, while advice is manually installed and
active until manually removed. It is not about the tool, but what you do with
the tool.

>> Here is another version on the same theme, where I don't think you could modify the local
>> environment without let-binding at all:
>>
>> #+begin_src emacs-lisp
>> (defun my-read-string (prompt)
>>   (let ((delta 20 )
>>         (minibuffer-mode-map org-mode-map))
>>     (window-resize (minibuffer-window) delta)
>>     (cl-letf (((symbol-function 'org-ctrl-c-ctrl-c)
>>                (lambda ()
>>                  (interactive)
>>                  (let ((s (buffer-string)))
>>                    (exit-minibuffer) s)))
>>               ((symbol-function 'minibuffer-mode) #'org-mode)
>>               ((symbol-function 'minibuffer-complete-and-exit) #'org-return)
>>               ((symbol-function 'org-kill-note-or-show-branches) #'keyboard-escape-quit))
>>       (read-string (concat "# Press C-c C-c to continue, C-c C-k to cancel\n# " prompt "\n\n")))))
>> #+end_src
>
> I hope I've convinced you to not do that. I definitely will not try it,
> as Emacs needs a working minibuffer for plenty of things: debugging,
> saving, quitting, etc.

Your minibuffer will continue to work during and after that function. If you
don't use recursive minibuffer, that will only affect the internal buffer
created by read-string.

If recursive minibuffer is enabled, yes they will also be affected. Of course,
there is no reason for you to try that, that was just an example, but I feel a
bit of passive aggressivity here, for no good reason tbh.

>> read-string is written in C and creates its own minibuffer, which is deleted by
>> the time read-string exits. I don't know of any other way to cutomize exactly
>> *that* minibuffer, without installing a hook or advising some functions, which I
>> think is way less clean and much more "global" than just running the function in
>> a local environment. As I understand, let binding for this purpose is a normal
>> technique in lisps, but I am not an expert as said; I am actually experimenting
>> with this for the purpose of learning and seeing what is possible.
>
> Yes, let binding is fundamental. But I think it's the first time I see
> 'cl-letf' with the 'symbol-function' place.

https://nullprogram.com/blog/2017/10/27/
https://endlessparentheses.com/understanding-letf-and-how-it-replaces-flet.html
https://stackoverflow.com/questions/39550578/in-emacs-what-is-the-difference-between-cl-flet-and-cl-letf

>> but I am not sure if I can do anything here without introducing at-least an
>> extra keymap, to not install into the org-capture-mode-map, so I can as well
>> create a minor mode, but at this point it is not much different than
>> re-invinting the read-string, so I'll terminate my experiment here :).
>
> You can replace the buffer keymap with a keymap that only contain your custom
> keys, and inherits everything else from org-capture-mode-map.

Isn't that what I wrote: introducing an extra keymap?

Of course I can solve the problem differently, but that was not what question
was about :).

>> But I wouldn't speak in some generic terms like "use hooks" or "advise" instead
>> of let-binding.
>
> I didn't mean to say to not use let bindings. I'm trying to say that
> using 'fset' (i.e. cl-letf + the symbol-function place) looks like a
> really bad idea to me.

Well, I definitely understand you, and agree that overwriting function for
everyone and everything is not the best idea, but unfortunately bindings work as
they do in Emacs. I would prefer to have a local binding, with cl-flet, but this
does not work in Emacs:

(defun my-read-string (prompt)
  (let ((delta 20 )
        (minibuffer-mode-map org-mode-map))
    (window-resize (minibuffer-window) delta)
    (cl-flet ((org-ctrl-c-ctrl-c ()
                (interactive)
                (let ((s (buffer-string)))
                  (exit-minibuffer) s))
              (minibuffer-mode () #'org-mode)
              (minibuffer-complete-and-exit () #'org-return)
              (org-kill-note-or-show-branches () #'keyboard-escape-quit))
      (read-string (concat "# Press C-c C-c to continue, C-c C-k to cancel\n# "
  prompt "\n\n")))))

>                        And, in this case, hooks and adivces are what is
> usually used.

Hooks serve a different purpose. Advice can serve same purpose with exactly
same side effect, and some other limitations. With some care, let-binding is
still more "local" then advice. With other words, I agree with you about the
problems, but not with dogmatic approach that it should never be done, and
that hooks and advices are the replacement.

>> (defun org-project-new-project ()
>>   (interactive)
>>   (let ((org-capture-templates org-project-templates))
>>     (org-capture)))
>>
>> ...
>
>> I don't know what would be the alternative, but let-binding on
>> org-capture-templates, let me clearly re-use the functionality without polluting
>> the global org-capture-templates variable and without re-implementing pretty
>> much anything.
>
> That looks to me like a perfect use of a let binding.  You bind a dynamic
> *variable* using *let*, and, that variable is rebound, and used, during, and
> only during 'org-capture'.
>
>
>
>> I am very interested to hear more on the topic, since I would definitely like to
>> learn more about different techniques.
>
> Variables are designed to be overriden (let bounds). Functions are not

I have never heard before that functions are not designed to be overriden. I
think of them as two slots in a symbol structure; let creates bindings for value
slot, and flet for function slot. Functions are just objects or data as any
other value in lisp.

> (as there is only one binding at any given time).

Yes, unfortunately, in Emacs it is so; but I don't think it should be :).

There is an interesting package by Nick Ferrier

https://github.com/nicferrier/emacs-noflet

but it does not seem to work, at least not for me.

regards
/a