Expose kernels to 3rd party extensions

[DonJayamanne]

This is a hackathon project, and the some of the original requirements can be found here #3998

Goal: Expose kernel to 3rd party extension authors so they can build on top of Jupyter extension

Possible extensions (prototyped):

• Other extensions can now execute code against kernels
  • User will be asked to allow such exensions to execute on behalf of them
  • 3rd party extensions can install missing packages, & the like (prototyped)
  • 3rd party extensions can inject kernel exensions (prototyped)
  • 3rd party extensions can intercept kernel execution requests
• SOS Jupyter Kernel (partially working)
  • We need cell pre-execute
• Support functionality found in Jupyter Sidecar widget
  • Allow user to open widgets outside the notebook.
• Custom variable explorer
  • Support for other languages (currently working, Python, R, Scala, C#, F#)
  • Support for variable visualizers (currently working for Python)
  • Other extensions can build on top and support more visualizers for each language
  • TOOD:
    • Provide extensibility for other extensions to provide details of each variable so that the tree view can be augmented (e.g. drilling down into details of each variable)
    • Ability to drill down into dict, dataframes in the variable viewer (partially protototyped)
• Better plot viewer
  • Custom visualizer for variable, to open a plot in a seprate view (use ipympl)
  • User doesn't need to isntall ipympl (source shipped with 3rd party extension)
• Custom data frameviewer
  • Basically similar to rest, display dataframe variable with the qgrid widget
  • Ship qgrid widget with 3rd party extension
  • Partially working
• Open altair (vega) plots with vega viewers
  • Convert vega into JSON, use 3rd party extensions to view the vega json data
  • I.e. more custom visualizers, just built on top of this new functionality

[DonJayamanne]

Todo:

• Publish sample extension(s)
  • https://github.com/microsoft/vscode-jupyter-powertoys
• Documentation (Wiki):
  • How to enroll into API
  • How to debug/test
  • Guidelines
• Release management
  • How do we notify changes (since we have a list of who can use this API, we can control this)
  • How do we introduce breaking changes into the API (stable/insiders releases)
    • Do we add entry into package.json->engine (our code can inspect callers package.json)
    • Do we publish an npm package that uses our API (the client package code can perform necessary checks)
  • E.g. what if user installs an extension that uses an old version of the API (do we fail silently, do we display a message, if so who - extension or us etc.)
• Publish *.d.ts files

[DonJayamanne]

Closing this for now, a sample usage of the API can be found here https://github.com/microsoft/vscode-jupyter-powertoys/