Building the website frontpage is necessary too.

pyvistaqt is on pypi! I can move forward with the migration

We have a few errors:

AttributeError: This plotter is closed and unable to save a screenshot.

UserWarning: VTK 9 no longer accepts an offset array

A crash:

―――――――――――――――――――――― mne/viz/_brain/tests/test_brain.py ――――――――――――――――――――――

[gw2] darwin -- Python 3.7.4 /Users/travis/miniconda/bin/python

worker 'gw2' crashed while running 'mne/viz/_brain/tests/test_brain.py::test_brain_timeviewer_traces[pyvista-both]'

A backend selection failure:

――――――――――――――――――――――――――――― test_get_3d_backend ――――――――――――――――――――――――――――――

[gw3] darwin -- Python 3.7.4 /Users/travis/miniconda/bin/python

mne/viz/backends/tests/test_renderer.py:162: in test_get_3d_backend

    assert renderer.get_3d_backend() == orig_backend

E   AssertionError: assert 'mayavi' == None

E    +  where 'mayavi' = <function get_3d_backend at 0x112db94d0>()

E    +    where <function get_3d_backend at 0x112db94d0> = <module 'mne.viz.backends.renderer' from '/Users/travis/build/mne-tools/mne-python/mne/viz/backends/renderer.py'>.get_3d_backend

Indeed the screenshot error is the one I got when I briefly tried to get it working. Not sure what the problem is there, but clearly just the from pyvistaqt import ... is not equivalent to what used to be done in pyvista :(

Let's try with BUILDING_GALLERY = True

Reference: pyvista/pyvista#783

I decided to change my approach. PyVista's architecture has changed a lot recently and this flow of issues is too much for me to handle at the same time. I will proceed from the latest working version, iterate one step at a time and adapt MNE to digest the changes.

Good start. master(78bc2a4) with 0.24.2 makes all the CIs happy and looking particularly at Azure here, I think it could be nice to pin to 0.24.2 instead of 0.24.0 as a short term solution @larsoner

Azure seems a tad slower but all good with 0.24.3 too, good news.

😭

(I speak for the CIs)

Not everything is green, it's not failed tests but timeouts.

The render window is cleared with _clear_ren_win() before the screenshot, that's why we get:

AttributeError: This plotter is closed and unable to save a screenshot.

Migrating completely to BackgroundPlotter should fix this one (work in progress in #7697).
Those PRs are oddly connected.

Feel free to combine into a single PR if you think it would help

And this one:

UserWarning: VTK 9 no longer accepts an offset array

has been fixed upstream normally (around pyvista 0.25.1)

On my macOS machine I changed environment.yml to have python<3.8, then ran:

conda env create -n pyvistaqt environment.yml
conda activate pyvistaqt
cd mne/viz/_brain/tests
jupyter notebook

Then ran stc.py, it worked. Jupyter notebook had a "unexpected keyword argument 'io_loop'" that I then fixed with pip install tornado==5.4.3. Then pip install pyimpl (probably should have used server_environment.yml. Then segfault in _sphere->_add_mesh->pyvista/plotting/plotting.py->core/filters.py line 3001 in compute_normals. So I think this is probably related to things not being properly initialized at the VTK end.

Many thanks! Your detailed report gives me some more material to work with to debug the situation.

Should we disable smooth_shading then as a short term solution?

Disabling for now to move forward. We can always enable it later once the situation is clarified.

Yes, that at least fixes test_notebook.py at my end!

Sadly, this fixes the segfault but the plots do not show brains (colorbars, etc. show up but not the brains) for me in the notebook

Any error log by any chance? What script did you use?

Never mind, actually I was testing in the wrong environment! Things work locally for me on macOS now so +1 for merge

Is it working for you too @agramfort?

@GuillaumeFavelier pushed 8be2205 to DRY our processEvents calls.

Wow, now it's packed with processEvents(). Fingers crossed for #7927

@GuillaumeFavelier my plan is to merge this pending @agramfort's approval, then #7931, then #7775, then #7927 (I'm about to enable traces there), probably with a merge or rebase at each step, sound good?

Sounds like a plan

works like a charm !

awesome work @GuillaumeFavelier and @larsoner !