search

enthought / mayavi

3D visualization of scientific data in Python
http://docs.enthought.com/mayavi/mayavi/
Other
1.3k stars 284 forks source link

Workarounds for VTK 9.1.0 #1097

Closed larsoner closed 2 years ago

larsoner commented 2 years ago

This PR is now ready to go with a few caveats. As commented by @prabhuramachandran in https://github.com/enthought/mayavi/pull/1097#issuecomment-965122185, we are observing segfaults when we test against vtk 9.1.0 from PyPI. See GitHub Actions CI output on commits https://github.com/enthought/mayavi/pull/1097/commits/a63b08a17c761e4e40fdc42040090356534e3773, https://github.com/enthought/mayavi/pull/1097/commits/a053a108a4b39cde7937f675a1265e75402fe29b or https://github.com/enthought/mayavi/pull/1097/commits/53345b160d54b6ef2771b7850e9b48e703c76652. The segfaults with the VTK 9.1.0 are happening on all platforms.

For now, we are choosing to ignore those failures and we are only testing against vtk 9.0.3 on CI but note that we have tested these changes locally on Mac OS and Windows.

Original PR description by @larsoner Depends/builds on #1050. Now I can at least get it to build+install on 9.1.0-rc2, but now I get: ``` mne/viz/tests/test_3d.py ..F >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>> traceback >>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>> mne/viz/tests/test_3d.py:125: in test_plot_sparse_source_estimates brain = plot_source_estimates( :24: in plot_source_estimates ??? mne/viz/_3d.py:1795: in plot_source_estimates return _plot_stc( mne/viz/_3d.py:1877: in _plot_stc brain = Brain(**kwargs) ../PySurfer/surfer/viz.py:488: in __init__ brain = _Hemisphere(subject_id, **kwargs) ../PySurfer/surfer/viz.py:3078: in __init__ self._geo_mesh = mlab.pipeline.triangular_mesh_source( ../mayavi/mayavi/tools/sources.py:1401: in triangular_mesh_source data_source.reset(x=x, y=y, z=z, triangles=triangles, scalars=scalars) ../mayavi/mayavi/tools/sources.py:831: in reset pd.polys = None E traits.trait_errors.TraitError: Cannot set the undefined 'polys' attribute of a 'PolyData' object. ``` Two issues I had to fix on top of #1050 1. `V.` was missing in the signatures, so I had to work around it 2. Arguments had type hints, so I had to sanitize those I'm not 100% sure I did either of these correctly, but again it at least let me build and try doing something, and even `import mayavi` (which indicates that `tvtk.Version.vtk_version` is exposed properly as a trait/property now). The fact that PolyData is broken is not a good sign, though. This is about all the time I can devote to this, it would be good if someone else could look and fix things for 9.1.0...
prabhuramachandran commented 2 years ago

Thank you, I will be taking a look at this tomorrow and later this week. Now that 9.1 is out, I want to fix this ASAP and push out a release that works with the latest VTK.

larsoner commented 2 years ago

Sounds good, feel free to take over or push commits here if you want @prabhuramachandran !

prabhuramachandran commented 2 years ago

@larsoner -- thanks, do you mind if I force-push once I make some headway? I rebased to master.

larsoner commented 2 years ago

Yes feel free

prabhuramachandran commented 2 years ago

So the tests are passing for me on macos with VTK 9.1 but there is some other issue on Linux with gh-actions, @rahulporuri can you please take a look. I hate that the tests stop when one architecture fails and do not have any more time this week.

prabhuramachandran commented 2 years ago

Looks like there are some major segfaults with VTK-9.1.0. The tests pass well with 9.0.3 and most of the time locally with 9.1.0 for me but on CI they segfault repeatably without any error message. Locally I get a traceback like this:

* thread #1, queue = 'com.apple.main-thread', stop reason = EXC_BAD_ACCESS (code=EXC_I386_GPFLT)
  * frame #0: 0x0000000165aa4c5f libvtkRenderingFreeType-9.1.dylib`vtkMathTextFreeTypeTextRenderer::SetScaleToPowerOfTwoInternal(bool) + 47
    frame #1: 0x0000000165b84c9f libvtkRenderingCore-9.1.dylib`vtkFlagpoleLabel::vtkFlagpoleLabel() + 559
    frame #2: 0x0000000165b83d04 libvtkRenderingCore-9.1.dylib`vtkFlagpoleLabel::New() + 52
    frame #3: 0x000000016c030668 libvtkRenderingVR-9.1.dylib`vtkOpenGLAvatar::vtkOpenGLAvatar() + 328
    frame #4: 0x000000016c0304dc libvtkRenderingVR-9.1.dylib`vtkOpenGLAvatar::New() + 28
    frame #5: 0x000000016499c811 libvtkWrappingPythonCore3.8-9.1.dylib`PyVTKObject_FromPointer + 609
    frame #6: 0x000000016499c599 libvtkWrappingPythonCore3.8-9.1.dylib`PyVTKObject_New + 153
    frame #7: 0x00000001000c3829 python`type_call + 41

So this does not seem to be mayavi specific although I need to investigate a little more. I think the build issues are fixed in this PR.

prabhuramachandran commented 2 years ago

@rahulporuri -- thanks, yes the summary seems accurate to me.

prabhuramachandran commented 2 years ago

Thanks. Merging for now, will look at the segfaults a little later.