Improve handling of context loss about k3d-jupyter

[martinal]

Doing a detach fixes the plot, and after reattach the plot is fine in the notebook too.

[martinal]

There's a (huge) branch of pythreejs which contains (besides basically a full rewrite) an approach to dealing with context loss, see this code:

jupyter-widgets/pythreejs@738f719#diff-2f0672887a2753687848ddc32e14306a

[artur-trzesiok]

It's very probably that in the latest version problem is solved. @martinal could you try to reproduce the issue ?

[martinal]

Issue is still there. To reproduce, create a new notebook, first cell:

import k3d

then N cells just displaying a plot each (for me N=17 triggered context loss):

k3d.plot()

Then when the browser console shows

WARNING: Too many active WebGL contexts. Oldest context will be lost.

scroll up to the top and the first plot looks like the attached picture.

[vidartf]

The pythreeJS solution to this is to keep N canvases around for rendering. Anything that tries to render something requests one of these, also supplying a callback for when it has to give it up. When it has to give it up, it replaces the canvas element with a PNG image of the last drawn frame.

Links to the latest code:

• The canvas manager is here: https://github.com/vidartf/pythreejs/blob/8c38c276bc4822e2f1dbb921ffd1ebc0b0a40d64/js/src/_base/RendererPool.js
• The code line to replace the current canvas with an image is here: https://github.com/vidartf/pythreejs/blob/23cad6ebdf297c4a58b888d54674b040814eb97d/js/src/_base/Renderable.js#L165