documentation not building

Describe the bug, what’s wrong, and what you expect:

I cannot get the documentation to build locally. The examples gallery builds fine and gets cached fine) but numpydoc seems to be blowing it up.

It helps dramatically to comment this line out:

https://github.com/pyvista/pyvista/blob/5dd060e465bd8339346de677d7b2b5aa661344af/doc/conf.py#L352

But it is still very slow during the “Reading sources…” steps.

Furthermore, when it does finish that, I am getting other errors which I’m currently tracking down

---

To Reproduce

Please include a code snippet to reproduce the bug in the block below:

cd doc
make html

---

System Information:

Python 3.8.12 | packaged by conda-forge | (default, Oct 12 2021, 21:50:38)
[Clang 11.1.0 ] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> import pyvista
>>> pyvista.Report()

--------------------------------------------------------------------------------
  Date: Thu Nov 25 22:45:22 2021 MST

                OS : Darwin
            CPU(s) : 12
           Machine : x86_64
      Architecture : 64bit
               RAM : 32.0 GiB
       Environment : Python
       File system : apfs
        GPU Vendor : ATI Technologies Inc.
      GPU Renderer : AMD Radeon Pro Vega 20 OpenGL Engine
       GPU Version : 4.1 ATI-4.7.29

  Python 3.8.12 | packaged by conda-forge | (default, Oct 12 2021, 21:50:38)
  [Clang 11.1.0 ]

           pyvista : 0.32.dev0
               vtk : 9.1.0
             numpy : 1.21.4
           imageio : 2.12.0
           appdirs : 1.4.4
            scooby : 0.5.9
            meshio : 4.4.6
        matplotlib : 3.5.0
           IPython : 7.29.0
          colorcet : 1.0.0
           cmocean : 2.0
        ipyvtklink : 0.2.1
             scipy : 1.7.3
              tqdm : 4.62.3
--------------------------------------------------------------------------------

[adeak]

We’ve sort of tracked the extract_edges issue down in slack, by which I just mean that it’s a VTK issue.

The following snippet crashes on VTK 9.1.0 and runs on VTK 9.0.3:

# dies on 9.1.0, works on 9.0.3                                                                          

from vtkmodules.vtkCommonComputationalGeometry import vtkParametricBoy                                   
from vtkmodules.vtkFiltersSources import vtkParametricFunctionSource                                     
from vtkmodules.vtkFiltersCore import vtkCleanPolyData, vtkFeatureEdges                                  
from vtkmodules.vtkFiltersModeling import vtkLinearSubdivisionFilter                                     
                                                                                                         
N_DIVISIONS = 3

# input: parametric Boy                                                                                  
para_function = vtkParametricBoy()
para_source = vtkParametricFunctionSource()                                                              
para_source.SetParametricFunction(para_function)
para_source.SetUResolution(100)
para_source.SetVResolution(100)
para_source.SetWResolution(100)
para_source.Update()
surf = para_source.GetOutput()                                                                           

# clean for manifoldness
clean = vtkCleanPolyData()
clean.SetPointMerging(True)
clean.SetConvertLinesToPoints(False)
clean.SetConvertPolysToLines(False)
clean.SetConvertStripsToPolys(False)
clean.SetInputData(surf)
clean.Update()
cleaned = clean.GetOutput()

# subdivide
sfilter = vtkLinearSubdivisionFilter()
sfilter.SetNumberOfSubdivisions(N_DIVISIONS)
sfilter.SetInputData(cleaned)
sfilter.Update()
subdivided = sfilter.GetOutput()

# extract feature edges
feature_edges = vtkFeatureEdges()
feature_edges.SetInputData(subdivided)
feature_edges.SetFeatureAngle(12)
feature_edges.Update()
mesh = feature_edges.GetOutput()

This is a simplified version of our Nefertiti edge extraction snippet. N_DIVISIONS has to be increased depending on available memory. At one point VTK 9.1.0 will crash with a huge amount of memory to be allocated (even though there’s sufficient memory for it), and 9.0.3 will run the same code just fine.

I haven’t raised this with VTK yet, I’m happy to hand over this opportunity.

[akaszynski]

This issue, along with several others (including build instability), is leading me to believe that we’re going to have to take some significant measures and build our own wheels.

I also get this issue when using the official vtk 9.1.0 wheel from pypi, but not when using my custom built wheel (compiled on Ubuntu 20.04) and not using manylinux2014. Nevermind, wheel works locally but not on our CI. Recommending modifying the example until this is fixed in VTK.