Re: window-text-pixel-size and presumable move_it_to change(s)

[Eli Zaretskii <eliz@gnu.org>]

> Date: Thu, 23 Jun 2016 10:18:20 +0200
> From: martin rudalics <rudalics@gmx.at>
> 
> ‘fit-window-to-buffer’ is currently broken on master when the window is
> horizontally combined.  I traced it back to the following differnece:
> Consider these two forms:
> 
> (window-text-pixel-size nil (window-start) (point-max) (frame-pixel-width) (window-body-height nil t))
> 
> (window-text-pixel-size nil (window-start) (point-max) nil (window-body-height nil t))
> 
> With emacs -Q on the release branch evaluating both forms via M-: gets
> me (576 . 48).
> 
> With emacs -Q on master evaluating the first form gets me (0 . 48) while
> evaluating the second form gets me (576 . 48).

Yes, I see something like that as well.

> I conclude that the behavior of move_it_to has changed recently when
> setting
> 
>        it.last_visible_x = max_x;
> 
> in Fwindow_text_pixel_size.  Is that conclusion correct?

No, the above still works as before.  The problem is in the new code
in Fwindow_text_pixel_size:

      x = min (move_it_to (&it, end, INT_MAX, max_y, -1,
			   MOVE_TO_POS | MOVE_TO_X | MOVE_TO_Y),
	       max_x);

This will call move_it_to twice, and the second call will return a
different value from the correct one, returned by the first call.
Replace this by something less fancy, like

      x = move_it_to (&it, end, INT_MAX, max_y, -1,
		      MOVE_TO_POS | MOVE_TO_X | MOVE_TO_Y);
      if (x > max_x)
        x = max_x;

and Bob'll be your uncle.

(In general, I suggest to make a rule to never use 'min' or 'max' when
their arguments are expressions.)