Improve the WebGL backend

[rossant]

This is a list of things to do to improve the WebGL backend once PR #592 is merged.

JavaScript loading

• Refactor vispy.js to use browserify instead of requireJS.
• Do not install nbextension if not necessary. The required JS files (notably vispy.min.js) are copied into the IPython profile everytime the ipynb_webgl backend is used. We could do a simple check to only copy if necessary (unless IPython's install_nbextension already does that).
• Fail gracefully if WebGL is not available (and suggest the user to use the VNC backend)

vispy.js

• Improve the GLIR implementation, notably:
  • in SHADERS, after attaching shaders, we should detach and destroy them
  • in TEXTURE, we should use the texture_number from the existing stored texture
  • try more complicated examples, notably with multiple textures
• Implement framebuffers in vispy.js (they are not supported currently).

Misc

• Solve issue Using initialize with gloo #606 so that all examples work in the notebook tackle issue Implement a server-side GLIR cache #615.
• Use binary WebSockets (implemented in IPython master and upcoming IPython 3.0) to transfer NumPy arrays from Python to JavaScript.
• Refactor ipynb_webgl so that other notebook backends can reuse most of the code (and bring back static and vnc).
• Rename paint into draw
• Implement VispyWidget(DOMWidget) (see comment below)

[mfkaptan]

Do not install nbextension if not necessary. The required JS files (notably vispy.min.js) are copied into the IPython profile everytime the ipynb_webgl backend is used. We could do a simple check to only copy if necessary (unless IPython's install_nbextension already does that).

I think install_nbextension already does that:
"By default, this compares modification time, and only stages files that need updating. If overwrite is specified, matching files are purged before proceeding."
There is also check_nbextension, which checks whether nbextension files have been installed. Do you think this should be used?

[rossant]

Hi @mfkaptan!
Yes we could use something like that. I might be wrong but it seemed to me that it wasn't updating all the time. Maybe we could have a force option in vispy.use_app('ipynb_***') that would force the reinstallation of the nbextension. That's useful for debugging. Or, if we had a general DEBUG option, it could always reinstall the nbextension in that case. @Eric89GXL WDYT?

[larsoner]

Could we add a function like vispy.app.backends.update_nbext() or so? Would that get you the debugging capabilities you want? We could also easily add a wrapper to make so you could do python make update_nbext from the vispy repo root.

[mfkaptan]

I have added force option as _prepare_js(force=False) to ipynb_webgl backend. I think the second part is to decide 'where should we get this option?' IMHO adding it as vispy.use_app('ipynnb_*, force=False') seems reasonable.

[rossant]

@mfkaptan I haven't tested it yet but that looks good to me.

[rossant]

Note to self:

I would like to make the following possible. I create a VisPy SceneCanvas as usual (e.g. a line plot). Next, I want to easily create an IPython DOMWidget that:

• exposes several trait attributes, for example line_color
• displays the SceneCanvas using WebGL in a cell's widget area

Here is how it could work:

Simple use

When there is no need to create a custom IPython widget.

import vispy
vispy.app.use_app('ipynb_webgl')

# create the SceneCanvas
canvas = ...

canvas.show()

Advanced use

import vispy
from vispy import VispyWidget
from IPython.display import display

# create the SceneCanvas
canvas = ...

class MyWidget(VispyWidget):
    """This class makes the bridge between the VisPy canvas and the WebGL widget.
    It exposes several trait attributes which notification callbacks can be customized.

    A VispyWidget is instanciated with a Canvas instance. Displaying the widget shows the
    WebGL version of the canvas.

    VispyWidget derives from IPython's DOMWidget.
    """

    # This trait attribute represent the line's color.
    line_color = 'blue'

    def on_trait_change(self, name, old, new):
        if name == 'line_color':
            self.canvas.myvisual.color = new

display(MyWidget(canvas))

[campagnola]

Isn't this already possible just by implementing properties on the subclass, like

@line_color.setter
def line_color(self, color):
    self.myvisual.color = color

?

[rossant]

Not exactly it really needs to be a trait attribute so that it works with IPython's MVC system.

[dragoljub]

This looks very exciting. Thanks and keep up the great work. 😄

[larsoner]

@dragoljub I think it's only supported on IPython version 3+, but that can be running Python 2.7 or 3.4 under the hood.

[dragoljub]

Yup I got it working after I updated to 4.0.

Thanks.

On Jun 1, 2015, at 7:00 AM, Eric Larson notifications@github.com wrote:

@dragoljub I think it's only supported on IPython version 3+, but that can be running Python 2.7 or 3.4 under the hood.

—
Reply to this email directly or view it on GitHub.

[djhoese]

I think besides possibly the "fail gracefully" task, I think I finished all of these tasks in the 0.6 release. I still consider the jupyter widget experimental (performance is not great), but it "works".