Zero height line causing arithmetic errors

Zero height line causing arithmetic errors

[Stefan Monnier]

My MPC.el package uses an overlay placed at the end of a line with an
after-string of #("\n" 0 1 (face (:height 0.05 :inverse-video t)))
in order to place a horizontal line in the display.

In Emacs-22, this worked fine (except the line was a bit thicker than
0.05 times the base line height).  In Emacs-CVS until recently this
worked except that the line was full-height (i.e. 13 pixels in my case).
Recently, it has started to cause more problems because now
FRAME_SMALLEST_FONT_HEIGHT returns 0, and that value is used at 2 places
in the divisor position.

I guess the 0 is because 13 pixels (default font height) * 0.05 -> 0.65
pixels which are truncated to 0.

I currently use the patch below, which makes it all work again (tho
still with the problem that the 0.05 isn't taken into account and line
line is 13 pixel think), but I suspect it's not the right place to fix
it,


        Stefan


=== modified file 'src/font.c'
--- src/font.c	2008-06-13 16:25:44 +0000
+++ src/font.c	2008-06-14 17:30:14 +0000
@@ -2682,6 +2682,7 @@
       if (FRAME_SMALLEST_FONT_HEIGHT (f) > font->height)
 	FRAME_SMALLEST_FONT_HEIGHT (f) = font->height, fonts_changed_p = 1;
     }
+  FRAME_SMALLEST_FONT_HEIGHT (f) = max (1, FRAME_SMALLEST_FONT_HEIGHT (f));
 #endif
 
   return font_object;

Re: Zero height line causing arithmetic errors

[Kenichi Handa]

In article <jwvtzfwc4kq.fsf-monnier+emacs@gnu.org>, Stefan Monnier <monnier@iro.umontreal.ca> writes:

> My MPC.el package uses an overlay placed at the end of a line with an
> after-string of #("\n" 0 1 (face (:height 0.05 :inverse-video t)))
> in order to place a horizontal line in the display.

> In Emacs-22, this worked fine (except the line was a bit thicker than
> 0.05 times the base line height).  In Emacs-CVS until recently this
> worked except that the line was full-height (i.e. 13 pixels in my case).
> Recently, it has started to cause more problems because now
> FRAME_SMALLEST_FONT_HEIGHT returns 0, and that value is used at 2 places
> in the divisor position.

> I guess the 0 is because 13 pixels (default font height) * 0.05 -> 0.65
> pixels which are truncated to 0.

> I currently use the patch below, which makes it all work again (tho
> still with the problem that the 0.05 isn't taken into account and line
> line is 13 pixel think), but I suspect it's not the right place to fix
> it,

For the problem of FRAME_SMALLEST_FONT_HEIGHT being set to
0, I installed the similar change.

And, for the problem of too tall line-height, I found the
culprit is XftTextExtents8.  When the pixelsize of a font is
less than 5 or so, for some font, it returns strange values
in `extents' argument.   For instance, with the attached
test problem, I get this result:

% ./xfttest 'bitstream vera sans mono'
pixelsize=9, ascent=8, descent=2
pixelsize=8, ascent=7, descent=2
pixelsize=7, ascent=6, descent=2
pixelsize=6, ascent=5, descent=2
pixelsize=5, ascent=4, descent=2
pixelsize=4, ascent=4, descent=1
pixelsize=3, ascent=3, descent=1
pixelsize=2, ascent=2, descent=1
pixelsize=1, ascent=1, descent=1
pixelsize=0, ascent=1, descent=1
% ./xfttest 'dejavu sans mono'
pixelsize=9, ascent=8, descent=2
pixelsize=8, ascent=7, descent=2
pixelsize=7, ascent=6, descent=2
pixelsize=6, ascent=5, descent=1
pixelsize=5, ascent=5, descent=1
pixelsize=4, ascent=5, descent=2  <-- increasing!
pixelsize=3, ascent=4, descent=4
pixelsize=2, ascent=5, descent=5
pixelsize=1, ascent=5, descent=5
pixelsize=0, ascent=5, descent=5

I don't know which component of these has a bug:
  o the font "dejavu sans mono",
  o freetype
  o xft

Anyway, I installed a quick dirty workaround; not believe
the result of XftTextExtents8 if the pixelsize is less than
5.

With that change, if your default font is scalable, the
following code works as expected.

(let ((buf (get-buffer-create "test"))
      overlay)
  (set-buffer buf)
  (erase-buffer)
  (insert "first line\n")
  (setq overlay (make-overlay (1- (point)) (point)))
  (overlay-put overlay 'after-string
	       (propertize "\n" 'face '(:height 0.01 :inverse-video t)))
  (switch-to-buffer buf))

But, note that the line height is still 2, not 1.  That's
because the display engine calculate the line height by
ascent+descent, and both ascent and descent doesn't become 0
in the structure XftFont even though `height' member is 1.
It may be another Xft (or freetype) problem.

---
Kenichi Handa
handa@ni.aist.go.jp

PS. You can compile xfttest.c by this command:
% gcc -g -o xfttest `xft-config --cflags` `xft-config --libs` xfttest.c

----- xfttest.c ----
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <X11/Xlib.h>
#include <X11/Xft/Xft.h>

char *str = "!\"#$%&'()*+,-./0123456789:;<=>?@ABCDEFGHIJKLMNOPQRSTUVWXYZ[\\]^_`abcdefghijklmnopqrstuvwxyz{|}~";

void
show_height (Display *display, XftFont *font)
{
  XGlyphInfo extents;

  XftTextExtents8 (display, font, str, 94, &extents);
  printf (" ascent=%d, descent=%d\n", extents.y, extents.height - extents.y);
}

int
main (int argc, char **argv)
{
  int size;
  char name[256], *p;
  Display *display;
      
  if (argc < 2)
    {
      fprintf (stderr, "Usage: xfttest FAMILYNAME\n");
      exit (1);
    }
  display = XOpenDisplay (NULL);
  if (! display)
    {
      fprintf (stderr, "Can't open the display\n");
      exit (1);
    }
  strcpy (name, argv[1]);
  p = name + strlen (name);
  for (size = 9; size >= 0; size--)
    {
      XftFont *font;

      sprintf (p, ":pixelsize=%d", size);
      font = XftFontOpenName (display, 0, name);
      printf ("pixelsize=%d,", size);
      show_height (display, font);
      XftFontClose (display, font);
    }
  
  exit (0);
}

Re: Zero height line causing arithmetic errors

[Dan Nicolaescu]

Kenichi Handa <handa@m17n.org> writes:

  > In article <jwvtzfwc4kq.fsf-monnier+emacs@gnu.org>, Stefan Monnier <monnier@iro.umontreal.ca> writes:
  > 
  > > My MPC.el package uses an overlay placed at the end of a line with an
  > > after-string of #("\n" 0 1 (face (:height 0.05 :inverse-video t)))
  > > in order to place a horizontal line in the display.
  > 
  > > In Emacs-22, this worked fine (except the line was a bit thicker than
  > > 0.05 times the base line height).  In Emacs-CVS until recently this
  > > worked except that the line was full-height (i.e. 13 pixels in my case).
  > > Recently, it has started to cause more problems because now
  > > FRAME_SMALLEST_FONT_HEIGHT returns 0, and that value is used at 2 places
  > > in the divisor position.
  > 
  > > I guess the 0 is because 13 pixels (default font height) * 0.05 -> 0.65
  > > pixels which are truncated to 0.
  > 
  > > I currently use the patch below, which makes it all work again (tho
  > > still with the problem that the 0.05 isn't taken into account and line
  > > line is 13 pixel think), but I suspect it's not the right place to fix
  > > it,
  > 
  > For the problem of FRAME_SMALLEST_FONT_HEIGHT being set to
  > 0, I installed the similar change.
  > 
  > And, for the problem of too tall line-height, I found the
  > culprit is XftTextExtents8.  When the pixelsize of a font is
  > less than 5 or so, for some font, it returns strange values
  > in `extents' argument.   For instance, with the attached
  > test problem, I get this result:
  > 
  > % ./xfttest 'bitstream vera sans mono'
  > pixelsize=9, ascent=8, descent=2
  > pixelsize=8, ascent=7, descent=2
  > pixelsize=7, ascent=6, descent=2
  > pixelsize=6, ascent=5, descent=2
  > pixelsize=5, ascent=4, descent=2
  > pixelsize=4, ascent=4, descent=1
  > pixelsize=3, ascent=3, descent=1
  > pixelsize=2, ascent=2, descent=1
  > pixelsize=1, ascent=1, descent=1
  > pixelsize=0, ascent=1, descent=1
  > % ./xfttest 'dejavu sans mono'
  > pixelsize=9, ascent=8, descent=2
  > pixelsize=8, ascent=7, descent=2
  > pixelsize=7, ascent=6, descent=2
  > pixelsize=6, ascent=5, descent=1
  > pixelsize=5, ascent=5, descent=1
  > pixelsize=4, ascent=5, descent=2  <-- increasing!
  > pixelsize=3, ascent=4, descent=4
  > pixelsize=2, ascent=5, descent=5
  > pixelsize=1, ascent=5, descent=5
  > pixelsize=0, ascent=5, descent=5

This works fine for me on Fedora 9:

./xfttest  'dejavu sans mono'
pixelsize=9, ascent=8, descent=2
pixelsize=8, ascent=8, descent=2
pixelsize=7, ascent=6, descent=2
pixelsize=6, ascent=5, descent=2
pixelsize=5, ascent=4, descent=2
pixelsize=4, ascent=4, descent=1
pixelsize=3, ascent=3, descent=1
pixelsize=2, ascent=2, descent=1
pixelsize=1, ascent=1, descent=1
pixelsize=0, ascent=1, descent=1

Re: Zero height line causing arithmetic errors

[Stefan Monnier]

> For the problem of FRAME_SMALLEST_FONT_HEIGHT being set to
> 0, I installed the similar change.

Thanks.

> And, for the problem of too tall line-height, I found the
> culprit is XftTextExtents8.  When the pixelsize of a font is

I think the problem I've been seeing is a different one.  While trying
to solve some other problem I discovered that my problem was linked to
issues where Xresource font settings seem impose themselves too
strongly, so the font-backend used was most likely X rather than xft (in
my case) and the problem disappeared if I removed those font settings
and let the font be specified some other way.  I.e. I'm glad you found
your bug, but mine is still around, and is probably
elsewhere altogether.

> But, note that the line height is still 2, not 1.  That's
> because the display engine calculate the line height by
> ascent+descent, and both ascent and descent doesn't become 0
> in the structure XftFont even though `height' member is 1.
> It may be another Xft (or freetype) problem.

Yes, I saw that.  I found it curious, but was actually happy about it:
I prefer my line to be 2 pixels thick (Emacs-22 makes it even thicker,
BTW).


        Stefan