Fix binary (un-)pickling of PyVista objects

Some time ago support was added for pickling/unplickling of PyVista dataobject. This feature is really useful in case of inter process communication like multi processing. However the feature is implemented use a ascii writer, which is tremendously slow.

I played somewhat with the code of @banesullivan (see: https://github.com/pyvista/pyvista-support/issues/59#issue-494762368) and tried to use the binary writer. I found out it was due to the pickle.dump the arrays were getting messed up. If you don’t pipe it to an output string but to a file the example works like a charm.

Of course some functionality has be added to removed the files after they are red but I would really like to see the SetFileTypeToBinary to be the default.

import vtk
import pickle
import pyvista as pv
import numpy as np

def pickle_vtk(mesh, filename):
    writer = vtk.vtkDataSetWriter()
    writer.SetInputDataObject(mesh)
    writer.SetFileName(filename)
    writer.SetFileTypeToBinary()
    writer.Write()
    return filename


def unpickle_vtk(filename):
    reader = vtk.vtkDataSetReader()
    reader.SetFileName(filename)
    reader.Update()
    return pv.wrap(reader.GetOutput())


#### Example

from pyvista import examples
mesh = examples.load_random_hills()

fname = 'filename.vtkpickle'
pickle_vtk(mesh, fname)
test = unpickle_vtk(fname)

assert np.allclose(mesh.points, test.points)
assert mesh.n_arrays == test.n_arrays

[MatthewFlamm]

I think @adeak comment could have separated better between the ask for more information and asking for better consideration of the people being pinged. This is my interpretation of the comment. Although I agree with both points separately.

The pinging seems nonspecific here and misses key contributors to this functionality. Large amounts of pings should be reserved for when there is a critical bug or a large blocker for continued development. This particular issue is localized.

Having two places for this conversation will be distracting, can we move this discussion to #3269 so that it is easier to track?

[MatthewFlamm]

We don’t even need to be able to cleanup the file if you use a BytesIO object, which I think should be pickle-able.

import vtk
import pickle
import pyvista as pv
import numpy as np
import io

def pickle_vtk(mesh, filename):
    writer = vtk.vtkDataSetWriter()
    writer.SetInputDataObject(mesh)
    writer.SetFileName(filename)
    writer.SetFileTypeToBinary()
    writer.Write()
    return filename


def unpickle_vtk(filename):
    reader = vtk.vtkDataSetReader()
    reader.SetFileName(filename)
    reader.Update()
    return pv.wrap(reader.GetOutput())


#### Example

from pyvista import examples
mesh = examples.load_random_hills()

fname = io.BytesIO()
pickle_vtk(mesh, fname)
test = unpickle_vtk(fname)

assert np.allclose(mesh.points, test.points)
assert mesh.n_arrays == test.n_arrays