[larsoner]

Make the "jupyterlite" backend an actual interactive backend named notebook_js. Performance is not great because on changes the whole thing needs to be re-serialized, but I'm going to work on streamlining that at the pyvista-js end. Changes ended up being pretty simple, with the exception being due to this re-serialization -- we need to "double buffer", i.e., render to a new widget fully before swapping, rather than a bare replace in order to avoid flashing. And we need some JS to make it clear an interactive update is in progress but waiting on the serialization. In the meantime zooming and rotating etc. is faster than the notebook backend (and the abstraction helps our code readability actually I think).

This code:

import mne

mne.viz.set_3d_backend("notebook_js")
data_path = mne.datasets.sample.data_path()
stc = mne.read_source_estimate(
    data_path / "MEG" / "sample" / "sample_audvis-meg", subject="sample"
)
brain = stc.plot(
    subjects_dir=data_path / "subjects",
    hemi="lh",
    views="lat",
    initial_time=0.1,
    time_viewer=True,
)

in a Jupyter notebook gives:

Also fixes a little double-event-publish bug along the way.

Changes drafted with Claude Opus 5.5 (especially helping with the JS!) but understood and iterated on by me.

[larsoner]

Converting to draft since tkoyama010/pyvista-js#634 will take updates from ~2.4s to ~0.4s, and simplify things a little bit for us (we won't need the double-buffering anymore!). We'll require quite a new pyvista-js (because that needs to be merged then a release made) but I think that's okay.

[larsoner]

Okay this is ready! Can be tested with the "launch lite" button for the dSPM example, which links here and for me gives:

[drammock]

@natinew77-creator do you want to review this one?

[natinew77-creator]

@natinew77-creator do you want to review this one?

Thanks for reaching out! I'd be happy to take a look. I'm currently busy with classes and midterm exams, but I'll review it as soon as I have some free time.