[vtk-developers] Need help with a crash on Kamino OSX Release mode.

Wed Apr 29 20:09:05 EDT 2015

Hi David,
   Thanks for this.
I have just added vtkScatterPlotMatrix along with vtkChartMatrix back,
oddly enough my windows machine never picked it up but one of the linux
buildbots did once. Then I took it out and everything seemed ok, it's back
in now.

I totally agree with you over the usefulness of Debug Visual Studio! It is
my main bug catcher. I use a little utility called Path Editor to reorder
the release and debug lib and bin paths in the environment variables
(PYTHONPATH and PATH).

Regards
    Andrew

On Wed, Apr 29, 2015 at 10:57 PM, David Cole <DLRdave at aol.com> wrote:

> The call stack for the out of range subscript assertion is:
>
>   msvcp120d.dll!00007fff4d2765d6() Unknown
>
> vtkChartsCore-6.3.dll!std::vector<vtkSmartPointer<vtkChart>,std::allocator<vtkSmartPointer<vtkChart>
> > >::operator[](unsigned __int64 _Pos) Line 1202 C++
>   vtkChartsCore-6.3.dll!vtkChartMatrix::GetChart(const vtkVector2i &
> position) Line 223 C++
>   vtkChartsCore-6.3.dll!vtkScatterPlotMatrix::SetActivePlot(const
> vtkVector2i & pos) Line 402 C++
> >
> vtkChartsCorePython27D-6.3.dll!PyvtkScatterPlotMatrix_SetActivePlot(_object
> * self, _object * args) Line 213 C++
>   python27.dll!000000001e0c2199() Unknown
>
> Looks like vtkScatterPlotMatrix_SetActivePlot can't be called from
> this test because other stuff that it depends on for proper operation
> is not usable as-initialized. This indicates to me that it may only be
> luck when it succeeds on non-VS-Debug builds, since the uninitialized
> variables, assuming that's what the cause is, will contain random
> contents when the test runs.
>
> When I add "vtkScatterPlotMatrix" as a Windows exception, the test
> passes on this build.
>
> I always love running Debug Visual Studio builds for catching easy
> errors like uninitialized variables and out of subscript references...
>
>
> HTH,
> D
>
>
> On Wed, Apr 29, 2015 at 8:38 AM, David Cole <DLRdave at aol.com> wrote:
> > I ran a build on my machine, named trivet, a Debug 64-bit build from
> > VS 2013, and the test fails with a single assertion failure and 10-20
> > mem leaks:
> >
> > https://open.cdash.org/testDetails.php?test=331705794&build=3791568
> >
> > Perhaps the "C:\Program Files (x86)\Microsoft Visual Studio
> > 12.0\VC\INCLUDE\vector(1201) : Assertion failed: vector subscript out
> > of range" report indicates the *real* cause of the problem on the Mac
> > Release build. I'm not sure which class/method that comes from, but
> > I'm about to run it and try to find out.
> >
> >
> > David C.
> >
> >
> >
> > On Wed, Apr 29, 2015 at 2:23 AM, Andrew Maclean
> > <andrew.amaclean at gmail.com> wrote:
> >> He Sean,
> >>
> >> I should clarify this a bit more.
> >> vtkDataEncoder invokes a thread pool which doesn't clean up after use,
> hence
> >> the crashes after program execution, adding vtkDataEncoder to the list
> of
> >> excluded classes "almost" fixed this.
> >> However Windows was still crashing when the program was run again, so,
> >> adding vtkWebApplication (which is the only class that calls
> >> vtkDataEncoder), to the list of excluded classes, fixed this issue.
> >>
> >> Notwithstanding this, it is puzzling, as I said in the previous e-mail
> as to
> >> why on Kamino we get a SIGSEGV at 0x0 address not mapped to object since
> >> vtkView is specifically excluded.
> >>
> >> Regards
> >>    Andrew
> >>
> >>
> >>
> >> On Wed, Apr 29, 2015 at 1:32 PM, Andrew Maclean <
> andrew.amaclean at gmail.com>
> >> wrote:
> >>>
> >>> Hi Sean,
> >>>
> >>> Yes, vtkDataEncoder and vtkWebApplication (which uses it) are excluded
> >>> form TestSetGet.py and TestEmptyInput.py. This fixed the crashes and
> also
> >>> the crash in windows when the program was run again. I think Windows
> was  a
> >>> bit slow recovering memory. after program execution.
> >>>
> >>> What is puzzling is that TestGetSet excludes vtkView so
> >>> vtkView.SetRepresentation() is never tested, this only fails on Kamino
> >>> (release). It passes on trey (osX release) and also on megas.
> >>>
> >>> Regards
> >>>    Andrew
> >>>
> >>>
> >>> On Wed, Apr 29, 2015 at 12:58 PM, Sean McBride <
> sean at rogue-research.com>
> >>> wrote:
> >>>>
> >>>> Did you fix the use-after-free we discussed on this list a few weeks
> ago?
> >>>>
> >>>> Sean
> >>>>
> >>>>
> >>>>
> >>>> On Wed, 29 Apr 2015 12:07:35 +1000, Andrew Maclean said:
> >>>>
> >>>> >David,
> >>>> >   Thanks for your insights.
> >>>> >Just finished testing this code on my Macbook Pro OS X Yosemite,
> Python
> >>>> >2.7.6 after building VTK in both Debug and Release mode. Everything
> runs
> >>>> > Ok
> >>>> >so I don't think it is a typedef or macro related to a release build
> on
> >>>> > OS
> >>>> >X.
> >>>> >
> >>>> >Has anyone else got any suggestions?
> >>>> >
> >>>> >
> >>>> >Regards
> >>>> >   Andrew
> >>>> >
> >>>> >On Wed, Apr 29, 2015 at 2:23 AM, David Gobbi <david.gobbi at gmail.com>
> >>>> > wrote:
> >>>> >
> >>>> >> Hi Andrew,
> >>>> >>
> >>>> >> The stack trace provides strong evidence that TestSetGet.py is not
> >>>> >> excluding vtkView on those dashboards, as ludicrous as that sounds.
> >>>> >>
> >>>> >>     c++filt -n _ZL27PyvtkView_SetRepresentationP7_objectS0_
> >>>> >>     PyvtkView_SetRepresentation(_object*, _object*)
> >>>> >>
> >>>> >> This method is only defined in vtkViewPython.cxx, and only
> >>>> >> dir(vtk.vtkView) will return this method.  Calling dir() on the
> >>>> >> vtkView
> >>>> >> subclasses will not return this method.  Hence, the only reason
> that
> >>>> >> TestSetGet.py would be calling this method is if it was testing
> >>>> >> vtkView.
> >>>> >>
> >>>> >> Perhaps someone added either a typedef or a macro that causes an
> >>>> >> additional reference to vtkView to appear in the vtk module under a
> >>>> >> different name.
> >>>> >>
> >>>> >>  - David
> >>>> >>
> >>>> >> On Tue, Apr 28, 2015 at 1:42 AM, Andrew Maclean
> >>>> >> <andrew.amaclean at gmail.com
> >>>> >> > wrote:
> >>>> >>
> >>>> >>> Hi All,
> >>>> >>>    I need some help here.
> >>>> >>>
> >>>> >>> I have converted TestSetGet.tcl to Python but I get a crash on the
> >>>> >>> dashboard for kamino building osx in release mode, here:
> >>>> >>> https://open.cdash.org/buildSummary.php?buildid=3789971
> >>>> >>> and here:
> >>>> >>> https://open.cdash.org/buildSummary.php?buildid=3789950
> >>>> >>>
> >>>> >>> In an earlier attempt, linux release on megas was also crashing in
> >>>> >>> release mode:
> >>>> >>>
> https://open.cdash.org/testDetails.php?test=331289192&build=3789418
> >>>> >>> Based on this I excluded vtkScatterPlotMatrix which fixed it on
> megas
> >>>> >>> but
> >>>> >>> not on Kamino.
> >>>> >>> So I excluded vtkChartMatrix and I still have the same errors on
> >>>> >>> Kamino.
> >>>> >>> But there are no clues as to where the crash is happening.
> However in
> >>>> >>> looking at:
> >>>> >>>
> https://open.cdash.org/testDetails.php?test=331378847&build=3789950
> >>>> >>> I think it is in vtkView - which is already excluded.
> >>>> >>> This could be the problem:
> >>>> >>> void SetRepresentation (vtkDataRepresentation *rep)
> >>>> >>> vtkDataRepresentation * GetRepresentation (int index=0)
> >>>> >>>
> >>>> >>> Can anyone help?
> >>>> >>>
> >>>> >>> I guess I can try excluding all the *View classes as documented
> in:
> >>>> >>> http://www.vtk.org/doc/nightly/html/classvtkView.html
> >>>> >>> Is there a less brute-force approach?
> >>>> >>>
> >>>> >>>
> >>>> >>> The code is here:
> >>>> >>> https://gitlab.kitware.com/vtk/vtk/merge_requests/151
> >>>> >>>
> >>>> >>> Thanks in advance for any help/guidance.
> >>>>
> >>>>
> >>>
> >>>
> >>>
> >>> --
> >>> ___________________________________________
> >>> Andrew J. P. Maclean
> >>>
> >>> ___________________________________________
> >>
> >>
> >>
> >>
> >> --
> >> ___________________________________________
> >> Andrew J. P. Maclean
> >>
> >> ___________________________________________
> >>
> >> _______________________________________________
> >> Powered by www.kitware.com
> >>
> >> Visit other Kitware open-source projects at
> >> http://www.kitware.com/opensource/opensource.html
> >>
> >> Search the list archives at:
> http://markmail.org/search/?q=vtk-developers
> >>
> >> Follow this link to subscribe/unsubscribe:
> >> http://public.kitware.com/mailman/listinfo/vtk-developers
> >>
> >>
>

-- 
___________________________________________
Andrew J. P. Maclean

___________________________________________
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/vtk-developers/attachments/20150430/15bc3f8e/attachment-0001.html>