Re: Control help- and Info-mode buffers from other buffers

[Juri Linkov <juri@linkov.net>]

>>> Anyway, pre/post hook hack is very useful, and works with many commands, but not
>>> with all, so it is not 100% failsafe and general.Try to execute Info-mode
>>> from other window but Info (shortcut 'm'). In my Emacs it does not work.
>>
>> I tried 'm' (Info-menu), and it works nicely without any problem.
>> As long as key prefixes are the same in both windows.
>
> Ok. Then perhaps I am doing something wrong? How do you run it, is the posted
> code snippet all you use, or is there something else there? For me it does not
> work at all.

With this code snippet:

#+begin_src emacs-lisp
(progn
  (add-hook 'pre-command-hook
            (lambda ()
              (other-window 1)
              (setq this-command (key-binding (this-command-keys)))))
  (add-hook 'post-command-hook (lambda () (other-window -1))))
#+end_src

in 'emacs -Q' type: C-x 4 4 C-h i m emacs RET

and it opens the Emacs Manual.

Then 'SPC SPC SPC DEL' scrolls it from the *scratch* buffer.

>> For the existing commands scroll-other-window, scroll-other-window-down,
>> recenter-other-window, beginning-of-buffer-other-window,
>> end-of-buffer-other-window, the user option that defines which window to use
>> is 'other-window-scroll-default', and it can be customized
>> to any function, for example, a function that looks for
>> a window with a Help/Info buffer on the current frame,
>> or on any other frame.  Or to a function that uses
>> 'get-mru-window' to get the most recently used/displayed window.
>> All this is customizable.
>
> Sure it is, but is isn't a customization problem. We wouldn't like to customize
> the stuff before every run, right? In a case like this, where we wish to run in
> a specific window like help, info or perhaps working-buffer window in case of
> ielm, we do want to make some specific commands, which means we would like to
> wrap that general do-in-X-window command. Otherwise it would be annoying to
> every time have to choose help window.

You can't avoid the need of customization even when using with-selected-window.
Since you have already seen requests to support renamed Help/Info buffers
like "*info*<2>", Man-mode buffer names like "*Man ...*", support frames
using the argument ALL-FRAMES of 'get-buffer-window', ...

> It would be a ginormous janitor work to go through all Emacs commands and
> re-write them. I don't think it is even possible. So no I don't suggest that
> :). I suggest this only for writing new commands, and I give a rough sketch as
> an illustration of what I men:
>
> #+begin_src emacs-lisp
> (defun test-command (arg &optional kill window)
>   (interactive "P")
>   (let* ((window-alist (mapcar (lambda (w) (cons (format "%s" w) w)) (window-list)))
>          (window
>           (cond
>            ((equal arg '(4))
>             (other-window-for-scrolling))
>            ((equal arg '(16))
>             (cdr (assoc (completing-read
>                          "Window: " window-alist) window-alist)))
>            (t (window-normalize-window window)))))
>     (with-selected-window window
>       (message "Did someting in window: %s" window))))
> #+end_src
>
> The let* wrapp could be generated on part of the user, in some way.
>
>> An alternative would be to put a new property on the command symbol
>> with a function that selects a window to redirect input to.
>
> How are that property and function meant to be implemented? By the end
> user, or by the Emacs?

Help/Info commands could have this property by default, then users could
add support for more commands by adding such a property to command symbols.

> Can end user just choose something like :run-in (one of nil, t,
> foo-mode, bar-mode, (some-predicate-p) some-function, etc), where "run-in" is
> the property you suggest, and the rest are constrains to choose from?

The property could define a function that selects a window
like in your code above.

> I don't tink it is too much different from what I suggest, tbh, since it will
> anyway have to select somehow the window and that selection would probably be
> steared by some argument to the command, but it will be coded differently.

Indeed, the implementation that selects a window could be the same.