to_vtk is too lenient with model array sizes

@thast pointed out to me that there is an issue when passing non-vectorized data arrays to the to_vtk() method.

import pyvista as pv
import numpy as np
import discretize

mesh = discretize.TensorMesh.readUBC('Mesh_model/mesh3d.msh')

model3d = np.load('Mesh_model/model.npy')
print(model3d.shape, model3d.size)

((10000, 28), 280000)

Then this:

mesh.to_vtk({'model3d': model3d})

works fine but as soon as you try to use that PyVista mesh, the VTK backend will segfault.

The reason you are able to create the PyVista object has to do with this check:

https://github.com/simpeg/discretize/blob/c16e34c93fd0b0b2848751ddde5cccbaadb5b82f/discretize/mixins/vtkModule.py#L88-L89

Should we change this to check the first axes of the model array? This way users could pass arrays containing vector data (nc by 3):

...
if mod.shape[0] != nc: 
     raise RuntimeError(...

[thast]

Thanks @banesullivan.

I think changing the test from size to shape is a good idea.

My issue when not vectorizing the model was that it did not throw me an error. Later, trying to use the pyvista object, I still did not get an error but my whole Python kernel died. This can be annoying if your script is long to run, and you do not know what went wrong. Getting an error earlier, without killing the whole kernel, would be very beneficial.

In most cases in SimPEG, we use order='F'. Non vectorized model are not common. In this case, it was just me making a silly oversight, while interpolating from multiple 1D mesh to a global 3D mesh.

[banesullivan]

Okay awesome! Then I think #163 is an appropriate fix for this issue and it should prevent kernel crashes.