Re: display word wrapping

[Juanma Barranquero <lektu@mi.madritel.es>]

On 30 May 2004 08:12:41 +0200, Eli Zaretskii <eliz@gnu.org> wrote:

> Does it crash (it _is_ a crash, isn't it?) under a debugger, or does
> that, too, cause the bug to change its behavior?

Yes, it is a crash. No, it fails the same way under debugging.

> If the former, run an unmodified optimized code under a debugger and
> use debugger facilities instead of debugging code to see what's wrong.

Yeah, I've already tried, and I see *where* it is failing, but debugging
optimized code is a bit of a pain.

> Comparison of machine code in the optimized and unoptimized versions
> might also tell you something useful.

It would, if I were more sure about what the code is exactly supposed to
do. I'm no expert in the image code.

> Finally, assuming that we are talking about a crash, posting the
> detailed description of the crash (exception number, register dump,
> stack dump, etc.) here could give some further ideas.

When I try to load a 811-byte .gif, I get:

  Unhandled exception at 0x0106dd08 in emacs.exe: 0xC0000005: Access
  violation reading location 0x00000004.

The stack trace is:

  image_spec_value(int spec=-536870912, int key=557610632, int * found=0x00000000) Line 931 + 0x8	C
  lookup_image(frame * f=0x015dbe00, int spec=-1572937664) Line 1682	C
  handle_single_display_prop(it * it=0x00000000, int prop=5, int object=-2109577216, text_pos * position=0x0082f04c, int display_replaced_before_p=0) Line 3702	C
  handle_display_prop(it * it=0x0082f04c) Line 3339 + 0x10	C
  handle_stop(it * it=0x0082efb0) Line 2591 + 0x3	C
  reseat(it * it=0x00000000, text_pos pos={...}, int force_p=1) Line 4676 + 0x6	C
  init_iterator(it * it=0x0082efb0, window * w=0x015dbe00, int charpos=1, int bytepos=1, glyph_row * row=0x02426800, face_id base_face_id=DEFAULT_FACE_ID) Line 2269 + 0x18	C
  start_display(it * it=0x0082efb0, window * w=0x0211ac00, text_pos pos={...}) Line 2288 + 0x1f	C
  try_window(int window=-2112771072, text_pos pos={...}) Line 12211 + 0x20	C
  redisplay_window(int window=-2112771072, int just_this_one_p=0) Line 11858 + 0xc	C
  redisplay_window_0(int window=-2112771072) Line 10588 + 0xa	C
  internal_condition_case_1(int (void)* bfun=0x0103fbcb, int arg=-2112771072, int handlers=-1589824248, int (void)* hfun=0x01029d11) Line 1374 + 0x35	C
  redisplay_windows(int window=-2112771072) Line 10569 + 0x1d	C
  redisplay_internal(int preserve_echo_area=0) Line 10152 + 0x8	C
  redisplay() Line 9386 + 0x7	C
  read_char(int commandflag=1, int nmaps=5, int * maps=0x0082fbec, int prev_event=557609984, int * used_mouse_menu=0x0082fc44) Line 2487	C
  read_key_sequence(int * keybuf=0x0082fce8, int bufsize=30, int prompt=557609984, int dont_downcase_last=0, int can_return_switch_frame=1, int fix_current_buffer=1) Line 8783 + 0x24	C
  command_loop_1() Line 1490 + 0x25	C
  internal_condition_case(int (void)* bfun=0x0105b380, int handlers=557688384, int (void)* hfun=0x01058003) Line 1334	C
  command_loop_2() Line 1271 + 0x15	C
  internal_catch(int tag=557684672, int (void)* func=0x0105c094, int arg=557609984) Line 1094 + 0x6	C
  command_loop() Line 1251	C
  recursive_edit_1() Line 961 + 0x5	C
  Frecursive_edit() Line 1023	C
  main() Line 1693 + 0x5	C
  mainCRTStartup() Line 259 + 0x12	C
  GetCurrentDirectoryW() + 0x44	

lookup_image is calling image_spec_value with the following code:

	  if (! img->background_valid)
	    {
	      bg = image_spec_value (img->spec, QCbackground, NULL);
	      if (!NILP (bg))
		{
		  img->background
		    = x_alloc_image_color (f, img, bg,
					   FRAME_BACKGROUND_PIXEL (f));
		  img->background_valid = 1;
		}
	    }

At that point, img->spec is not valid. Previously, img has been
initialized from spec (the arg passed to the function), so spec and
img->spec should be equal, but they aren't anymore. If you change the
call to do

 bg = image_spec_value (spec, QCbackground, NULL);

it succeeds, but then the call to postprocess_image

	  if (!EQ (*img->type->type, Qpostscript))
	    postprocess_image (f, img);

fails for the same reason (img->spec is not valid).

On optimized code, img is not on the stack, but on a register. In fact,
if I change

  struct image *img;

to

  static struct image *img;

Emacs works fine. So it looks like it *is* really a bug in the optimizer
code, which is clobbering a register or something like that.


                                                           /L/e/k/t/u