WIP: Port 3d testing to BackgroundPlotter completely

[GuillaumeFavelier]

This PR ports testing of the PyVista 3d backend to the BackgroundPlotter completely.

As a consequence, the variable MNE_3D_BACKEND_TESTING needs refactoring.

Known issue

• The dimensions of a screenshot taken with the BackgroundPlotter are not equal to win_height x win_width x 3 but slightly less. I suppose that the real dimensions taken into account are effectively those of the render view (which is smaller than the window).

[GuillaumeFavelier]

I reported this behaviour in https://github.com/pyvista/pyvista/issues/703

[GuillaumeFavelier]

One way to mitigate this issue is to use setMinimumSize() on the render view widget so that the window does not automatically shrink it.

[GuillaumeFavelier]

the window does not automatically shrink it.

The drawback is that the user is no longer able to resize the window to a geometry smaller then window_size.

[GuillaumeFavelier]

This can be relevant to #7211 (about minimum/fixed widget geometry)

[larsoner]

The drawback is that the user is no longer able to resize the window to a geometry smaller then window_size.

Can you set this until the super init finishes, then un-set it?

[GuillaumeFavelier]

I use it after BackgroundPlotter init to resize the render view. I'll see if there is a way to un-set it without updating the geometry of the window.

[larsoner]

I use it after BackgroundPlotter init to resize the render view.

Can you see if it's possible to do before initializing the view? It should help prevent more jerky resizes, if those happen with the current approach

[GuillaumeFavelier]

The window is resized before being shown so I don't think this will lead to jerkiness. But this is a workaround without a way to setting it at init. So far, locally the main issue I notice is with show_traces=True. When the window opens:

I have to manually resize to get all the content:

Also, the CIs do not seem to like those changes.

[larsoner]

Probably whatever setMinSize you do for the 3D area, you need to do for the 2D area as well. Maybe at 1/4 of whatever the requested 3D vertical size is?

[larsoner]

For now you could assert_allclose(shape, expected_shape, atol=40) # XXX undo once size fixed or so to get things to pass. Seems worth it to get the actual BackgroundPlotter used, then we can deal with fixing the size separately

[GuillaumeFavelier]

Awesome idea, thank you. I was stuck with SetMinimumSize() tbh

[larsoner]

The Azure failure is fixed by #7740, but I couldn't see the failed OSX log so I restarted that build.

[larsoner]

Failures on OSX apparently seem real

https://travis-ci.org/github/mne-tools/mne-python/builds/683971748

[GuillaumeFavelier]

Actually if #7758 is chosen, I would prefer to keep testing on both plotters.