> Date: Sun, 4 Nov 2012 18:44:51 +0200
> From: eliz@gnu.org
> Subject: Re: bug#12774: 24.2.50; EMACS 24.2.50.1 crashing‏
> To: vincent.b.1@hotmail.fr
> CC: 12774@debbugs.gnu.org
> 
> > From: Vincent Belaïche <vincent.b.1@hotmail.fr>
> > Date: Sat, 3 Nov 2012 23:24:15 +0100
> > 
> > The crash can happen without my doing anything special previously
> > except opening that attached file USING_CEDET_FROM_BZR which causes
> > loading orgmode. Also it seems that it has to do with the size of my
> > init.el file --- and not only to its content.
> 
> OK. Here's what I'm asking you to do. Start Emacs under GDB, like
> this:
> 
> c:\Documents and Settings\Vincent>gdb C:/Programme/GNU/Emacs/bin/emacs.exe
> 
> (Note: do NOT invoke Emacs via the --exec= switch, since that does not
> load the symbol table and other debugging info from emacs.exe.)
> 
> Once GDB shows its prompt, type these commands:
> 
> (gdb) break ralloc.c:353
> (gdb) break w32heap.c:141
> (gdb) break w32heap.c:155
> (gdb) run
> 
> Please verify that the 3 source lines where you set breakpoints are
> the ones marked below:
> 

My ralloc.c is Bzr-110429 and the breakpoint line is 352 instead of 353

> ralloc.c:
> 
> if ((*real_morecore) (- excess) == 0)
> {
> /* If the system didn't want that much memory back, adjust
> the end of the last heap to reflect that. This can occur
> if break_value is still within the original data segment. */
> >>>>>>> last_heap->end = (char *) last_heap->end + excess;
> /* Make sure that the result of the adjustment is accurate.
> It should be, for the else clause above; the other case,
> which returns the entire last heap to the system, seems
> unlikely to trigger this mode of failure. */
> if (last_heap->end != (*real_morecore) (0))
> emacs_abort ();
> }
> 

My w32heap.c is Bzr-110429, and the breakpoint lines are 194 instead of
141, and 208 instead of 155.

> w32heap.c:
> 
> /* Sanity checks. */
> if ((data_region_end - size) < data_region_base)
> >>>>> return NULL;
> 
> /* We can only decommit full pages, so allow for
> partial deallocation [cga]. */
> new_data_region_end = (data_region_end - size);
> new_data_region_end = (unsigned char *)
> ((DWORD_PTR) (new_data_region_end + syspage_mask) & ~syspage_mask);
> new_size = real_data_region_end - new_data_region_end;
> real_data_region_end = new_data_region_end;
> if (new_size > 0)
> {
> /* Decommit size bytes from the end of the heap. */
> if (using_dynamic_heap
> && !VirtualFree (real_data_region_end, new_size, MEM_DECOMMIT))
> >>>>> return NULL;
> }
> 
> If your sources are different, adjust the line numbers in the "break"
> commands accordingly.
> 
> Now do whatever you usually do to get Emacs to crash, and wait for one
> of the breakpoints to break (they will break before the crash, so
> don't expect the abort dialog).
> 
> Then please tell which of the breakpoints on the 2 "return NULL" in
> w32heap.c breaks. If it's the second one, the one after the call to
> VirtualFree, please type this command:
> 
> (gdb) print w32_last-error()
> 

The break occurred in the 3rd breakpoint (second one in w32heap.c),
i.e. *NOT* the one after the call to VirtualFree, but one the previous
one in the same file.

> (note the parentheses: they are important) and tell what it prints.
> 
> Then please type "continue", which should stop at the breakpoint in
> ralloc.c, and tell what do the following commands produce:
> 
> (gdb) print last_heap->end
> (gdb) print last_heap->bloc_start
> (gdb) print excess
> (gdb) print sbrk(0)
> 
> Also, please type "xbacktrace" and post the Lisp-level backtrace this
> produces. If you don't invoke GDB from the src directory, you may
> need to type "source /path/to/emacs/src/.gdbinit" to make "xbacktrace"
> work.
>

Please find attached the log which I got.

Please note the following:

- at line 83 the message `Undefined command: "xbacktrace".  Try "help".'
  occurs again because I typed RET a second time.

- at line 89 `Are you sure you want to change it? (y or n) [answered Y;
  input not from terminal]' the answered was autnomously produced by
  EMACS which I am was using as a terminal --- because that makes it
  easier to save the log to a file and to recall past commands --- I
  don't know why EMACS used with `M-x shell' does not always behave well
  when you are asked interactive questions, but that is a completely
  different issue.

- At line 109 I tried print w32_last-error() to see what happens, even
  though that is not the 2nd breakpoint (1st in w32heap.c)

- At lines 111 & 112 I typed RET twice, and this is why you get the same
  error messages again `No symbol "w32_last" in current context.'

> 
> Thanks.

All thanks are to you for your kind support.

   Vincent.