[vtk-developers] londinium/debug dynamic library issues

[Marcus D. Hanwell]

On Mon, Feb 4, 2013 at 12:24 PM, David Gobbi <david.gobbi at gmail.com> wrote:
> Hi All,
>
> I've been investigating why the londinium debug/coverage build is
> failing so many tests.  The failures appear to have begun on Oct 12:
> http://open.cdash.org/viewTest.php?onlyfailed&buildid=2608121
>
> Tcl tests failed with this error:
>  dlopen: cannot load any more object with static TLS
>
> A few Python tests failed with this error:
>  Inconsistency detected by ld.so: dl-tls.c: 422:
> _dl_allocate_tls_init: Assertion `map->l_tls_modid == cnt' failed!
>
> The Python tests that failed with this error were as follows:
>  vtkChartsCorePython-TestBarGraph
>  vtkChartsCorePython-TestLinePlot
>  vtkChartsCorePython-TestLinePlotColors
>  vtkChartsCorePython-TestParallelCoordinatesColors
>  vtkChartsCorePython-TestScatterPlotColors
>  vtkChartsCorePython-TestStackedPlot
>  vtkCommonComputationalGeometryPython-TestParametricFunctions
>  vtkFiltersModelingPython-TestNamedColorsIntegration
>  vtkRenderingTkPython-TestTkRenderWidget
>  vtkRenderingTkPython-TestTkRenderWindowInteractor
>
> Since only a few Python tests failed while others succeeded, I'm sure
> that the contents of these failing tests can provide some clues as to
> why these TLS (thread-local storage) errors are occurring when Python
> and Tcl try to dlopen certain VTK libraries on londinium/debug.
>
> The commit that caused the problem was merged sometime between Oct 7
> and Oct 12 (VTK was suffering from compile errors that week, so it
> can't be narrowed down any further than that).
>
I was traveling when you posted this, and now it seems that the CDash
entries have been deleted (I can't go back that far). I just updated
the packages on the machine, and am rerunning the tests now. It seems
like Tcl is in a bad state on that machine (I maintain it). Looking
into it, I have a very similar machine that acts as my main
development system that suffers from none of these failures.

It may be some ccache issue (perhaps I should not use ccache for
nightly builds as it seems to have a non-zero number of subtle bugs
like this). Looking into it (and seeing the same tests pass on its
sister system, which is why I suspect some local cruft is causing the
issue). I will clear ccache and see what tonight's nightly reports.

Marcus