bug#47300: delete-window to select window with same position

[Juri Linkov]

> +      (when (and (>= x (nth 0 edges)) (< x (nth 2 edges))
> +                 (>= y (nth 1 edges)) (< y (nth 3 edges)))
>
> We have to mention the "<" somewhere.  For two side-by-side windows with
> the right window's point at 1, we obviously must avoid the left window's
> right edge because the X, Y coordinates set by pos_visible_p are the top
> left edge of the character box at that position, i.e., 0, 0.
>
> But OTOH the "<" leaves the right edge of a frame's rightmost window
> unhandled and someone might use that function for other purposes.

I changed <= to < to fix only Y coordinates for top/bottom windows,
but maybe X coordinates should still use <= ?

> It would be a tad more elegant to just save `posn-at-point' here ...
>
> +              (setq posn-at-point-x (+ (nth 0 edges) (car posn-at-point))
> +                    posn-at-point-y (+ (nth 1 edges) (cdr posn-at-point))))))
>       ;; Actually delete WINDOW.
>       (delete-window-internal window)
>       (window--pixel-to-total frame horizontal)
> -     (when (and frame-selected
>
> ... and delay extracting its car and cdr until here.
>
> +        (when (and posn-at-point-x posn-at-point-y
> +                (setq window-at-posn-at-point
> +                      (window-at-pos posn-at-point-x posn-at-point-y frame)))

Please note that the crucial difference is that now it uses
frame relative positions rather than window relative ones
with '(window-edges window nil nil t)' of the window to be deleted,
and window coordinates added to X and Y of posn-at-point.

> I'd still urge you to transfer the `get-mru-window' call here too, that
> is, the entire
>
>         /* Now look whether `get-mru-window' gets us something.  */
>         mru_window = call1 (Qget_mru_window, frame);
>         if (WINDOW_LIVE_P (mru_window)
>             && EQ (XWINDOW (mru_window)->frame, frame))
>           new_selected_window = mru_window;
>
>         /* If all ended up well, we now promote the mru window.  */
>         if (EQ (FRAME_SELECTED_WINDOW (f), selected_window))
>           Fselect_window (new_selected_window, Qnil);
>         else
>           fset_selected_window (f, new_selected_window);
>
> block, and to include the
>
>         (when (and frame-selected
>                  (window-parameter
>                   (frame-selected-window frame) 'no-other-window))
>         ;; `delete-window-internal' has selected a window that should
>         ;; not be selected, fix this here.
>         (other-window -1 frame))
>
> so we can avoid selecting a new window up to four times when deleting a
> window (the one remaining call in `delete-window-internal' does not
> record the window and has to stay there).  It's simply not nice to do
> that.

1. `get-mru-window' could be one possible choice of the new option.
2. `use-posn-at-point' could be another choice.
3. Selecting the first window could remain in `delete-window-internal'.
4. But what to do with this code block with '(other-window -1 frame)'?
   Should this be simply deleted?