Notebook communication between JavaScript and Python

[Conchylicultor]

With Colab/Jupyter, it is possible to have bidirectional communication between Python/Javascript, without installing a notebook extention:

• In Jupyter, it is possible to execute arbitrary Python code from JS:

  IPython.notebook.kernel.execute('x = 2');

• In Colab, for security reason, it is forbidden. However, it is still possible to call explicitly registered Python functions with input/outputs serialised to Json.

  import IPython.display
  from google.colab import output
  
  def my_func():
  print('called')
  
  output.register_callback('my_func', my_func)
  
  IPython.display.Javascript("""
  google.colab.kernel.invokeFunction('my_func', [], {});  // fn_name, args, kwargs
  """)

It is very convenient, and unlock powerful use-cases. For example:

• pandas.DataFrame in Colab can be viewed interactively (with filters, pages,...).
• ecolab.inspect(obj) to interactivelly inspect arbitrary Python objects: https://www.youtube.com/watch?v=MFbb5wZlYvE

There's an duplicate issue which was closed (https://github.com/microsoft/vscode-jupyter/issues/6918), even though the original issue (https://github.com/microsoft/vscode-jupyter/issues/7551) hasn't been fixed.

• If this is possible, it would be great to have an example on how to do this.
• If not, it would be great if VSCode was exposing an API VSCode.kernel.invokeFunction similar to Colab.

[slundberg]

Would also love this. Just noting that this is already possible, just not supported (#13243). Which means allowing it does not change the current level of security.

[DonJayamanne]

Closing this issue, as the package https://github.com/manzt/anywidget now makes it very easy for front end code to communicate with the kernels using ipywidgets.

Though jupyter ntoeb9ook exposes a way to communicate with the kernel, jupyter lab no longer supports this. I believe using IPyWidgets is the right way (as thats what Jupyter lab also recomends)