Mayavi / VTK

[ArneVoss]

Describe the bug

Mayavi 4.8.3 is incompatible with numpy > 2.4.0

Mayavi tries to import Numpy's function 'in1d', which has been deprecated since NumPy 2.0 and is removed in favor of numpy.isin (https://numpy.org/doc/stable/release/2.4.0-notes.html)

Therefore, pytest fails with an AttributeError: module 'numpy' has no attribute 'in1d'

Current situation

Limit numpy to the last compatible version (numpy<2.4.0), see PR #80

Long-term solution

Although Mayavi was the best solution at the time I created the GUIs, it is now getting old and there is little to no development (https://github.com/enthought/mayavi/graphs/code-frequency).

1. Replace Mayavi, for example with pyvista.
   New strategy: Keep Mayavi for now but switch to PySide6.
2. At the same time, think about replacing pyface (also by Enthought) with pyside as QT binding. That is cleaner and should help to reduce problems caused with Qt during installation.

Implemented with PR #83.

[ArneVoss]

Update: after a closer look, it's not Mayavi, but VTK itself!

    from vtk.numpy_interface import algorithms as algs
/usr/share/miniconda/envs/pytest_env/lib/python3.12/site-packages/vtkmodules/numpy_interface/algorithms.py:982: in <module>
    in1d = _make_ufunc(numpy.in1d)

Help is under way: Kitware/VTK@0f9b0ce

The above-mentioned fix is in v9.6.0.rc2, so it should be included in version 9.6.0.

[ArneVoss]

Update: VTK 9.6 is out since a couple of weeks. However, is asks for Python 3.14, which is a step I don't want to take now, because other tools I use (e.g. flowsim) aren't so fast with updating to new Python versions. That means I'm stuck here for a while...

[ArneVoss]

Update: VTK 9.6.2 support multiple Python versions, but now Mayavi has not (yet) been updated to a newer VTK version and still requires VTK 9.4. In addition, there is nearly no maintenance of Mayavi, so pyvista becomes more interesing again.