bug#24804: 25.1; posn-at-point erroneously signals an error

[Eli Zaretskii <eliz@gnu.org>]

> Date: Fri, 28 Oct 2016 11:12:19 +0300
> From: Eli Zaretskii <eliz@gnu.org>
> Cc: 24804@debbugs.gnu.org, politza@hochschule-trier.de
> 
> > Thanks.  But according to Eli the problem is in Fpos_visible_in_window_p
> > which should never return a negative y in the first place.
> 
> Maybe it could, when an image is vscrolled?
> 
> In any case, the proposed patch is IMO incorrect: when point is at a
> large image (or even a character glyph that's very tall relative to
> its window's height), this patch will pretend that point is invisible,
> whereas in fact it's partially visible (as is clear from the use case
> discussed here).
> 
> The key to understanding why the patch is wrong is to remember that
> the Y coordinate of a glyph on display is the coordinate of the top of
> the glyph.  For a large glyph, its top can be off-screen, but most of
> it could still be visible.  This is evident in the value returned by
> pos-visible-in-window-p in this case: (747 -36 36 6 925 0).  It
> exactly means that the image starts 36 pixels above the window top
> (and ends 6 pixels below its bottom).
> 
> Perhaps if RTOP is non-nil, we should pass Y + RTOP to posn-at-x-y.

Andreas, can you try the patch below?  Thanks.

diff --git a/src/keyboard.c b/src/keyboard.c
index bb411e7..65938a5 100644
--- a/src/keyboard.c
+++ b/src/keyboard.c
@@ -10790,11 +10790,19 @@ The `posn-' functions access elements of such lists.  */)
     {
       Lisp_Object x = XCAR (tem);
       Lisp_Object y = XCAR (XCDR (tem));
+      Lisp_Object aux_info = XCDR (XCDR (tem));
+      int y_coord = XINT (y);
 
       /* Point invisible due to hscrolling?  X can be -1 when a
 	 newline in a R2L line overflows into the left fringe.  */
       if (XINT (x) < -1)
 	return Qnil;
+      if (!NILP (aux_info) && y_coord < 0)
+	{
+	  int rtop = XINT (XCAR (aux_info));
+
+	  y = make_number (y_coord + rtop);
+	}
       tem = Fposn_at_x_y (x, y, window, Qnil);
     }