Memory sharing issues after PyVista 0.37 for RectilinearGrid

[banesullivan]

See 0e64e7d

pyvista/pyvista#3179 broke the test_convert_vtr test when it validates that the X and Y coordinates share memory but not Z. This is super odd and I cannot figure out what is going wrong as the changes in pyvista/pyvista#3179 do not seemingly affect this. So for some reason X and Y are DEEP copied now while Z is still SHALLOW copied.

@MatthewFlamm, I hate to throw you in the deep end but I'm wondering if you might have any insights come to mind as to why that pull request would have broken this test.

[banesullivan]

the reality is that a lot of the code here in pyvista-xarray for ensuring shared arrays is very fragile and needs major improvement

[MatthewFlamm]

This is weird. This test does not pass in a vtk object at all to Rectilinear Grid as far as I can tell. One line reads in a file, the other instantiates a blank object and then manually constructs it via attributes here

pyvista-xarray/pvxarray/rectilinear.py

Lines 24 to 29 in 0e64e7d

	if x is not None:
	self._mesh.x = self._get_array(x)
	if y is not None:
	self._mesh.y = self._get_array(y)
	if z is not None:
	self._mesh.z = self._get_array(z)

Other weird thing is that the prior behavior for Rectilinear grid was to always deep copy vtk objects.

[MatthewFlamm]

Have you bisected it to this exact commit in PyVista? Or just by version?

[banesullivan]

I bisected, and it was the commit for pyvista/pyvista#3179