Yeah, this solves my problem! Thank you!

this PR breaks the screenshots for me on my mac. I get screenshots like this now

Thanks for pointing it out @agramfort because test_brain_screenshot is skipped for the macos/conda job:

mne/viz/_brain/tests/test_brain.py::test_brain_screenshot[pyvista] SKIPPED [ 95%]
...
SKIPPED [2] mne/viz/_brain/tests/test_brain.py:351: Unreliable/segfault on macOS CI

https://github.com/mne-tools/mne-python/pull/9351/checks?check_run_id=2440366851#step:12:2839
https://github.com/mne-tools/mne-python/pull/9351/checks?check_run_id=2440366851#step:12:3040

@larsoner what do you think about trying the alternative calculation only when the reshape fails?

Thanks for pointing it out @agramfort because test_brain_screenshot is skipped for the macos/conda job:

I think it's just a 0-1 vs 0-255 mapping problem, I'll test locally and push a fix

@larsoner what do you think about trying the alternative calculation only when the reshape fails?

I'd rather not have two code paths. If we're forced to go back to raw mode, we need to figure out somehow what shape the output actually needs to be. But like I mention above I doubt it will be necessary

This code:

import os.path as op
import mne
from mne.minimum_norm import read_inverse_operator, apply_inverse

# Set dir
data_path = mne.datasets.sample.data_path()
subject = 'sample'
data_dir = op.join(data_path, 'MEG', subject)
subjects_dir = op.join(data_path, 'subjects')
fname_evoked = data_dir + '/sample_audvis-ave.fif'
condition = 'Left Auditory'
evoked = mne.read_evokeds(fname_evoked, condition=condition,
                          baseline=(None, 0))
inv = read_inverse_operator(
    data_path + '/MEG/sample/sample_audvis-meg-oct-6-meg-inv.fif')
stc_vec = apply_inverse(evoked, inv, pick_ori='vector')
with mne.viz.use_3d_backend('pyvista'):
    brain = stc_vec.plot(hemi='both', views=['lat', 'med'],
                         initial_time=0.1, subjects_dir=subjects_dir)
import matplotlib.pyplot as plt
plt.figimage(brain.screenshot(time_viewer=True), resize=True)

Gives on main:

On what @agramfort tested you can see two bugs -- one I introduced and another involving bgcolor that is on main but not seen there in this example (white line running vertically along the lower left part of the image):

Pushed a commit to fix both problems:

Clearly the image of the brain is wrong, and the matplotlib plot is not the right size either. But these are separate issues on main, too, so let's fix them separately.

@GuillaumeFavelier do you want to try my code snippet to see if you see the same problems on your system?

Okay the problem was really with my figimage call, I needed resize=True (edited now above) and I get something correct:

But there is a problem with the update of the display in that the views don't respond to the fact that the matplotlib traces have forced them to resize:

I would have to switch to KDE for a better test but on i3 (which maximizes the new windows by default) I think I can reproduce:

When I leave the maximized mode:

Could it be yet another process events?

For me, deleting the self._interactor.resize(ren_sz.width(), ren_sz.height()) in _window_ensure_minimum_sizes makes it look correct. Going to see what it breaks on Linux and macOS locally

... actually it doesn't make it work, it makes it look not cut off, which is good, but it's the wrong size, which is bad. I'll see if I can figure out what to do to make it work

Okay should be fixed, can you review and test locally @GuillaumeFavelier ?

Thanks @larsoner!!