boolean_union produces a non-manifold mesh

Describe the bug, what’s wrong, and what you expected.

boolean_union produces non-manifold meshes, preventing further processing operations.

Steps to reproduce the bug.

import pyvista

sphere_center = (2.4623486548662186, 15.274176120758057, 2.042705088853836)
cone_center = (1.7672341, 16.30967, 1.2092979)
cone_direction = (-0.4634096920490265, 0.6903295516967773, -0.5556047558784485)

cone = pyvista.Cone(center=cone_center, direction=cone_direction, height=3, radius=1, resolution=8).triangulate()
sphere = pyvista.Sphere(center=sphere_center, radius=1).triangulate()
cylinder = pyvista.Cylinder(center=(sphere_center[0], sphere_center[1]/2, sphere_center[2]), direction=(0, 1, 0), radius=0.5, height=sphere_center[1]).triangulate()

result = sphere.boolean_union(cylinder).triangulate().boolean_union(cone).clean()

result.clip_closed_surface((0, 1, 0), (0, 15.274176120758057, 0)) # This fails due to a non-manifold mesh

Saving the result shows that it is, indeed, non-manifold. See the screenshots section.

System Information

--------------------------------------------------------------------------------
  Date: Tue Jun 14 12:19:43 2022 BST

                OS : Linux
            CPU(s) : 24
           Machine : x86_64
      Architecture : 64bit
       Environment : Python
       GPU Details : error

  Python 3.9.11 (main, Mar 17 2022, 07:20:01)  [GCC 11.2.0]

           pyvista : 0.34.1
               vtk : 9.1.0
             numpy : 1.21.5
           imageio : 2.19.2
           appdirs : 1.4.4
            scooby : 0.5.12
        matplotlib : 3.5.1
             PyQt5 : Version unknown
             scipy : 1.7.3
--------------------------------------------------------------------------------

Screenshots

The result of the above code is this non-manifold mesh:

A triangle that should be cut off at the end isn’t, which makes it cut into the interior of the mesh.

Code of Conduct

• I agree to follow this project’s Code of Conduct

[whophil]

That’s awesome! Will it be possible to install alongside conda-forge vtk/pyvista, or does vtk itself need a custom re-compile?

It installs alongside vtk from conda-forge (it just links against the VTK libs at compile-time).

[darikg]

I am working on packaging vtkbool for conda on the conda-forge channel. If you are interested in this, you can follow the progress here zippy84/vtkbool#51

That’s awesome! Will it be possible to install alongside conda-forge vtk/pyvista, or does vtk itself need a custom re-compile?

For my application, vtkbool worked better, as cell/point data of the input meshes were passed through without the need for additional bookkeeping. I am sure this can be achieved with the CGAL bindings, but I didn’t have time to look into it.

Yeah the trade-off CGAL has is you can handle cell/point data however you want but you have to implement it yourself by subclassing PMP::Corefinement::Default_visitor. I have an example here for keeping track of which of the two original meshes the corefined face originates from, but I’m not good enough at C++ / too lazy to make it work for all properties in general.

Also of interest is this recent arxiv paper; looks like the authors will be releasing the source code when the paper is published.