Re: EXC_BAD_ACCESS [was: persistent listeners]

[Ben Abbott]

On Mar 13, 2011, at 4:54 AM, Søren Hauberg wrote:

> lør, 12 03 2011 kl. 23:06 +0100, skrev logari81:
>> On Sat, 2011-03-12 at 16:42 -0500, Ben Abbott wrote:
>>> On Mar 12, 2011, at 4:26 PM, logari81 wrote:
>>> 
>>>> On Sat, 2011-03-12 at 16:14 -0500, Ben Abbott wrote:
>>>>> On Mar 12, 2011, at 12:55 PM, Ben Abbott wrote:
>>>>> 
>>>>>> On Mar 12, 2011, at 11:59 AM, Ben Abbott wrote:
>>>>>> 
>>>>>>> On Mar 12, 2011, at 3:10 AM, logari81 wrote:
>>>>>>> 
>>>>>>>> On Thu, 2011-03-10 at 21:20 +0000, bpabbott wrote:
>>>>>>>>> On Mar 10, 2011, at 02:34 PM, Konstantinos Poulios
>>>>>>>>> <address@hidden> wrote:
>>>>>>>>> 
>>>>>>>>>> On Wed, Mar 9, 2011 at 12:16 PM, Konstantinos Poulios
>>>>>>>>>> <address@hidden> wrote:
>>>>>>>>>>> In the new implementation of subplot we have the problem that the
>>>>>>>>>>> listeners that we use for the synchronization of subplots are
>>>>>>>>>> deleted
>>>>>>>>>>> once something is plotted on a subplot.
>>>>>>>>>>> 
>>>>>>>>>>> The attached patches implement persistent listeners, i.e.:
>>>>>>>>>>> 
>>>>>>>>>>> 1. listeners added with:
>>>>>>>>>>> addlistener (handle, propertyname, cbfunction, "persistent")
>>>>>>>>>>> will not be removed by the c++ method "remove_all_listeners", nor
>>>>>>>>>> by a
>>>>>>>>>>> call to delete_listener() with empty arguments list.
>>>>>>>>>>> 
>>>>>>>>>>> 2. however, explicitly removing a persistent listener from the
>>>>>>>>>> frontend with:
>>>>>>>>>>> dellistener (handle, propertyname, cbfunction)
>>>>>>>>>>> should still work.
>>>>>>>>>>> 
>>>>>>>>>>> Please take a look at the patches so that I can push them if it is
>>>>>>>>>> ok,
>>>>>>>>>>> so that we can make some progress with the subplot fixes.
>>>>>>>>>>> 
>>>>>>>>>>> Kostas
>>>>>>>>>> 
>>>>>>>>>> I have just pushed a modified/fixed version of these patches.
>>>>>>>>>> Actually
>>>>>>>>>> with a bit modified policy than the one described in my previous
>>>>>>>>>> mail:
>>>>>>>>>> 
>>>>>>>>>> 1. One can add a persistent listener with
>>>>>>>>>> addlistener (handle, propertyname, cbfunction, "persistent")
>>>>>>>>>> 
>>>>>>>>>> 2 One can delete all persistent listeners on a property with
>>>>>>>>>> dellistener (handle, propertyname, "persistent")
>>>>>>>>>> 
>>>>>>>>>> This change will allow us to fix the known issues with subplot.
>>>>>>>>>> 
>>>>>>>>>> Regards
>>>>>>>>>> 
>>>>>>>>>> Kostas
>>>>>>>>> 
>>>>>>>>> With this in place, Octave crashes for me when I try to create a plot
>>>>>>>>> with fltk active.
>>>>>>>>> 
>>>>>>>>> 
>>>>>>>>> octave:1> graphics_toolkit fltk
>>>>>>>>> 
>>>>>>>>> Reading symbols for shared libraries . done
>>>>>>>>> 
>>>>>>>>> Reading symbols for shared libraries . done
>>>>>>>>> 
>>>>>>>>> octave:2> plot (1:10)
>>>>>>>>> 
>>>>>>>>> Reading symbols for shared libraries . done
>>>>>>>>> 
>>>>>>>>> 
>>>>>>>>> Program received signal EXC_BAD_ACCESS, Could not access memory.
>>>>>>>>> 
>>>>>>>>> Reason: KERN_PROTECTION_FAILURE at address: 0x00000000
>>>>>>>>> 
>>>>>>>>> 0x927d725e in glMatrixMode ()
>>>>>>>>> 
>>>>>>>>> (gdb) bt
>>>>>>>>> 
>>>>>>>>> #0  0x927d725e in glMatrixMode ()
>>>>>>>>> 
>>>>>>>>> #1  0x03705dbe in OpenGL_fltk::setup_viewport ()
>>>>>>>>> at 
>>>>>>>>> /Users/bpabbott/Development/mercurial/octave/src/DLD-FUNCTIONS/__init_fltk__.cc:146
>>>>>>>>> 
>>>>>>>>> #2  0x03705dbe in OpenGL_fltk::resize (this=0x3609a50, xx=0, yy=20,
>>>>>>>>> ww=640, hh=440) at DLD-FUNCTIONS/__init_fltk__.cc:129
>>>>>>>>> 
>>>>>>>>> #3  0x037155d6 in fltk_uimenu::show ()
>>>>>>>>> at 
>>>>>>>>> /Users/bpabbott/Development/mercurial/octave/src/DLD-FUNCTIONS/__init_fltk__.cc:788
>>>>>>>>> 
>>>>>>>>> #4  0x037155d6 in plot_window::show_menubar ()
>>>>>>>>> at 
>>>>>>>>> /Users/bpabbott/Development/mercurial/octave/src/DLD-FUNCTIONS/__init_fltk__.cc:789
>>>>>>>>> 
>>>>>>>>> #5  0x037155d6 in plot_window::plot_window (this=0x36094d0, xx=0,
>>>>>>>>> yy=0, ww=640, hh=480, address@hidden) at
>>>>>>>>> DLD-FUNCTIONS/__init_fltk__.cc:737
>>>>>>>>> 
>>>>>>>>> #6  0x03715caa in plot_window [inlined] ()
>>>>>>>>> at 
>>>>>>>>> /Users/bpabbott/Development/mercurial/octave/src/DLD-FUNCTIONS/__init_fltk__.cc:755
>>>>>>>>> 
>>>>>>>>> #7  0x03715caa in figure_manager::do_new_window (this=0x33b8870,
>>>>>>>>> address@hidden) at DLD-FUNCTIONS/__init_fltk__.cc:1572
>>>>>>>>> 
>>>>>>>>> #8  0x037042bb in figure_manager::new_window ()
>>>>>>>>> at 
>>>>>>>>> /Users/bpabbott/Development/mercurial/octave/src/DLD-FUNCTIONS/__init_fltk__.cc:1450
>>>>>>>>> 
>>>>>>>>> #9  0x037042bb in __fltk_redraw__ () at
>>>>>>>>> DLD-FUNCTIONS/__init_fltk__.cc:1767
>>>>>>>>> 
>>>>>>>>> #10 0x03715f77 in fltk_graphics_toolkit::redraw_figure
>>>>>>>>> (this=0x3606c40, address@hidden) at
>>>>>>>>> DLD-FUNCTIONS/__init_fltk__.cc:1865
>>>>>>>>> 
>>>>>>>>> #11 0x001fdef4 in graphics_toolkit::redraw_figure ()
>>>>>>>>> at /Users/bpabbott/Development/mercurial/octave/src/graphics.h:7618
>>>>>>>>> 
>>>>>>>>> #12 0x001fdef4 in Fdrawnow (address@hidden) at graphics.cc:7618
>>>>>>>>> 
>>>>>>>>> #13 0x00498fad in octave_builtin::do_multi_index_op (this=0x3864140,
>>>>>>>>> nargout=0, address@hidden, lvalue_list=0x0) at ov-builtin.cc:128
>>>>>>>>> 
>>>>>>>>> #14 0x00499377 in octave_builtin::do_multi_index_op (this=0x3864140,
>>>>>>>>> nargout=0, address@hidden) at ov-builtin.cc:98
>>>>>>>>> 
>>>>>>>>> #15 0x0007e578 in octave_value::do_multi_index_op (this=0xbfffb518,
>>>>>>>>> nargout=0, address@hidden) at ov.cc:1266
>>>>>>>>> 
>>>>>>>>> #16 0x0005d5f1 in feval (address@hidden, address@hidden,
>>>>>>>>> nargout=0) at oct-parse.yy:4140
>>>>>>>>> 
>>>>>>>>> #17 0x0003af99 in interactive_input (address@hidden,
>>>>>>>>> force_readline=false) at input.cc:237
>>>>>>>>> 
>>>>>>>>> #18 0x0003ba25 in std::string::operator= () at /usr/include/c
>>>>>>>>> ++/4.2.1/bits/basic_string.h:280
>>>>>>>>> 
>>>>>>>>> #19 0x0003ba25 in octave_gets [inlined] ()
>>>>>>>>> at /Users/bpabbott/Development/mercurial/octave/src/input.cc:280
>>>>>>>>> 
>>>>>>>>> #20 get_user_input [inlined] ()
>>>>>>>>> at /Users/bpabbott/Development/mercurial/octave/src/input.cc:351
>>>>>>>>> 
>>>>>>>>> #21 0x0003ba25 in octave_read (buf=0x3a50600 "", max_size=8192) at
>>>>>>>>> input.cc:373
>>>>>>>>> 
>>>>>>>>> #22 0x0003ccd3 in yy_get_next_buffer () at lex.cc:3097
>>>>>>>>> 
>>>>>>>>> #23 0x0004df11 in octave_lex () at lex.cc:2939
>>>>>>>>> 
>>>>>>>>> #24 0x00063a9d in octave_parse () at oct-parse.cc:2395
>>>>>>>>> 
>>>>>>>>> #25 0x0041a5c9 in main_loop () at toplev.cc:567
>>>>>>>>> 
>>>>>>>>> #26 0x003b591d in octave_main (argc=6, argv=0xbfffc354, embedded=0) at
>>>>>>>>> octave.cc:935
>>>>>>>>> 
>>>>>>>>> #27 0x00001f80 in main (argc=6, argv=0xbfffc354) at main.c:35
>>>>>>>>> 
>>>>>>>>> 
>>>>>>>>> Occasionally, I see some strange results on MacOS. Can someone else
>>>>>>>>> confirm this behavior?
>>>>>>>>> 
>>>>>>>>> Ben
>>>>>>>> 
>>>>>>>> Hmm, this is very strange because normal plotting shouldn't involve any
>>>>>>>> listeners at all, should it? I couldn't reproduce this issue on two
>>>>>>>> different computers but I cannot try it on a Mac.
>>>>>>>> 
>>>>>>>> Have you tried rebuilding everything from scratch?
>>>>>>>> 
>>>>>>>> Kostas
>>>>>>> 
>>>>>>> I don't see this on Ubuntu, but am able to reproduce in on two Macs. 
>>>>>>> The root of the problem may be due to fltk-aqua-1.1.10 , the older gcc 
>>>>>>> 4.2.1 I'm using, or Apple's patches to gcc. 
>>>>>>> 
>>>>>>> Ben
>>>>>> 
>>>>>> Kostas,
>>>>>> 
>>>>>> I just confirmed your changeset (12505) is not the cause of the problem. 
>>>>>> After reverting it, I still get a core-dump.
>>>>>> 
>>>>>> I'll try to isolate the problem.
>>>>>> 
>>>>>> Ben
>>>>> 
>>>>> I've isolated the problem to the change below.
>>>>> 
>>>>> http://hg.savannah.gnu.org/hgweb/octave/diff/9c831d357e6f/src/DLD-FUNCTIONS/__init_fltk__.cc
>>>>> 
>>>>> Without this change, I don't get the crash.
>>>>> 
>>>>> Ben
>>>> 
>>>> aha, this patch consists of two parts. Could you first apply only the
>>>> changes below: @@ -776,28 +780,28 @@
>>>> and see if this triggers the issue. If not, then we will have isolated
>>>> the problem in the first part.
>>>> 
>>>> I have also cc'd the original author of __init_fltk__.cc, so that he
>>>> maybe can review my patch.
>>>> 
>>>> Kostas
>>> 
>>> I had the same thought. After a few experiments, I've isolated the problem. 
>>> Does this change interfere with the intent of your changeset?
>>> 
>>> Ben
>> 
>> Yes your last change would cancel my intention in my previous change.
>> The idea is that the lines:
>> 
>> ||    if (uimenu->items_to_show ())
>> ||      show_menubar ();
>> ||    else
>> ||      hide_menubar ();
>> 
>> should be before the first allocation of the canvas:
>> 
>> ||      // This allows us to have a valid OpenGL context right away.
>> ||      canvas->mode (FL_DEPTH | FL_DOUBLE );
>> ||      if (fp.is_visible ())
>> ||        {
>> ||          show ();
>> ||          if (fp.get_currentaxes ().ok())
>> ||            show_canvas ();
>> ||          else
>> ||            hide_canvas ();
>> ||        }
>> 
>> otherwise, at least with my fltk version, the first time anything is
>> plotted is vertically shifted by the menu bar height which was very
>> annoying.
>> 
>> So we have to find some other solution. Maybe Shai could give us a hint.
> 
> I haven't really been following this thread, so I might be way of
> here... but it sounds like you're re-introducing this bug:
> 
> https://savannah.gnu.org/bugs/?31931
> 
> Maybe you should look at how this was fixed?
> 
> Søren

Nice catch.

I can confirm that with the change modified so that fltk functions on MacOS, 
the command below core-dumps

        figure (1, "visible", "off")

Ben