Why boolean operations with this meshes are empty?

[vargonis]

I have two example meshes for which pyvista fails to correctly compute the difference. I thus decided to try with meshlib, but the difference (as well as the union!) comes out empty. Here's the code:

from meshlib import mrmeshpy as mm, mrmeshnumpy as mn
import pyvista as pv

# Conversion utilities:
def meshlib_to_pyvista(mesh: mm.Mesh):
    points = [list(p) for p in mesh.points]
    faces = [x for f in mesh.topology.getAllTriVerts() for x in [3, *[v.get() for v in f]]]
    return pv.PolyData(points, faces)

def pyvista_to_meshlib(polydata: pv.PolyData):
    verts = polydata.points
    faces = polydata.faces.reshape(-1, 4)[:,1:]
    return mn.meshFromFacesVerts(faces, verts)

# The meshes:
line1 = pv.Line(pointa=(0, 0, 0), pointb=(1, 0, 0))
arc1 = pv.CircularArc(pointa=(1, 0, 0), pointb=(0, 1, 0), center=(1, 1, 0), resolution=50)
line2 = pv.Line(pointa=(0, 1, 0), pointb=(0, 0, 0))
boundary = line1 + arc1 + line2
boundary.faces = [len(boundary.points), *range(len(boundary.points))]
x = boundary.extrude([0, 0, 1], capping=True).triangulate()
c = pv.Cube().triangulate()

# Sanity check: both look good
meshlib_to_pyvista(pyvista_to_meshlib(c)).plot()
meshlib_to_pyvista(pyvista_to_meshlib(x)).plot()
(
    meshlib_to_pyvista(pyvista_to_meshlib(x))
    + meshlib_to_pyvista(pyvista_to_meshlib(c))
).plot() # ...and they have nontrivial intersection and symmetric difference.

# Union calculation (I was interested in the difference, but this failure is even more puzzling)
union = mm.boolean(
    pyvista_to_meshlib(x),
    pyvista_to_meshlib(c),
    mm.BooleanOperation.Union,
).mesh
meshlib_to_pyvista(union) # fails, because diff is empty: it has neither vertices nor faces!
list(union.points) # []

What's going on? Is this a bug, or I'm doing something wrong? Here's an image of the meshes:

[Fedr]

Please note that mm.boolean returns empty mesh in case of error. Please check for the error as follows:

booleanRes = mm.boolean(
    pyvista_to_meshlib(x),
    pyvista_to_meshlib(c),
    mm.BooleanOperation.Union,
)
if not booleanRes.valid():
    print(booleanRes.errorString)
union = booleanRes.mesh

Hopefully, the error message will be clear. If not, please attach your input meshes for us.

[vargonis]

There is an error indeed: "Cannot separate mesh B to inside and outside parts, probably contours on mesh B are not closed or are not consistent." The mesh generated by vtk is pretty bad, but the countour seems closed and consistent:

I guess I should either let meshlib generate the mesh, or apply some mesh improvement algorithm. Any suggestions? Thanks!

[Fedr]

Could you please attach both your meshes (e.g. as .PLY files) that you pass into Boolean? Then we will be able to explain in detail what is wrong with them and give some advice.

[vargonis]

Here they are. Mesh "a.ply" is the problem, it has degenerate triangles (as you can see from the wireframe I posted above). I just made a quick search for "meshlib remesh" but didn't find anything relevant.

meshes.zip

[Grantim]

Hello @vargonis!

It turned out that pyvista does not create consistent faces' indices order by default, and really mesh looks like this:

(you can see that faces at top are inverted (actually, they are OK but all others are inverted according to ply files that you've provided))

To find it out I saved the mesh in MehsLib internal format like this;

mm.saveMesh(pyvista_to_meshlib(x),"a.mrmesh")

---

I tried to have a look at normals with pyvista (with plot_normals) and it seemed OK, but actually it was not, pyvista fixes normals automatically for visualization

(all normals are looking inward on the image)

So, in this case you need to fix faces orientation before converting to meshlib like this:

x = boundary.extrude([0, 0, 1], capping=True).triangulate().compute_normals(consistent_normals=True)

compute_normals as it fixes faces orientation.

After this fix boolean works OK.

[vargonis]

Great, thanks a lot, it works! So, by consistency here it was meant consistent orientation. I had also plotted the normals with pyvista, so It didn't occur to me that could be the problem.