GNU bug report logs - #78444
30.1; Crash in GC (vector_marked_p)

Previous Next

Full log

Message #83 received at 78444 <at> debbugs.gnu.org (full text, mbox):

From: Pip Cet <pipcet <at> protonmail.com>
To: George P <georgepanagopo <at> gmail.com>
Cc: martin rudalics <rudalics <at> gmx.at>, Eli Zaretskii <eliz <at> gnu.org>,
 acorallo <at> gnu.org, 78444 <at> debbugs.gnu.org
Subject: Re: bug#78444: 30.1; Crash in GC (vector_marked_p)
Date: Sun, 15 Jun 2025 17:49:42 +0000

"George P" <georgepanagopo <at> gmail.com> writes:

> Yes, that is precisely what it says....
>
> (gdb) p displayed_buffer
> $1 = (struct buffer *) 0xfffffffffffffffb

That's XBUFFER (Qnil), possibly set by this code in redisplay_internal:

  else if (FRAME_REDISPLAY_P (sf) && !FRAME_OBSCURED_P (sf))
    {
      sf->inhibit_clear_image_cache = true;
      displayed_buffer = XBUFFER (XWINDOW (selected_window)->contents);
      /* Use list_of_error, not Qerror, so that
	 we catch only errors and don't run the debugger.  */
      internal_condition_case_1 (redisplay_window_1, selected_window,
				 list_of_error,
				 redisplay_window_error);

However, I believe that's only the surface of the bug: the selected
window's contents should never be nil, right?  According to this comment
in window.h, that would mean it's a pseudo window, but another comment
says the selected window is always a leaf window:

    /* For a leaf window or a tooltip window this is the buffer shown
       in the window; for a combination window this is the first of
       its child windows; for a pseudo window showing the menu bar or
       tool bar this is nil.  It is a buffer for a minibuffer window
       as well.  */
    Lisp_Object contents;

/* This is the window in which the terminal's cursor should be left when
   nothing is being done with it.  This must always be a leaf window, and its
   buffer is selected by the top level editing loop at the end of each command.

   This value is always the same as FRAME_SELECTED_WINDOW (selected_frame).  */

So this looks like an inconsistent state reached by the redisplay
machinery.  We'd have to check whether it's the code above or actually
this code:

      if (update_miniwindow_p)
	{
	  Lisp_Object mini_window = FRAME_MINIBUF_WINDOW (sf);

	  displayed_buffer = XBUFFER (XWINDOW (mini_window)->contents);
	  internal_condition_case_1 (redisplay_window_1, mini_window,
				     list_of_error,
				     redisplay_window_error);
	}

While it would probably would have avoided the crash to check whether
XWINDOW (whichever)->contents is Qnil before setting displayed_buffer,
I don't think that's the right fix: violating that internal redisplay
assumption is a bad thing, and we need to figure out why it can happen
when windows are being deleted (possibly in delete-window-internal?)

Can you look up the line numbers against your source tree to see which
code segment was active when the error happened?

Thanks!
Pip

Previous Next

GNU bug tracking system
Copyright (C) 1999 Darren O. Benham, 1997,2003 nCipher Corporation Ltd, 1994-97 Ian Jackson.