Support Jupyter Notebook Interactive Computing paradigm

[FuriouslyCurious]

Around 500,000 VSCode users have been relying on Don Jayamanne has "vscodeJupyter" extension to connect to Jupyter Notebooks and execute code in an interactive environment. (https://github.com/DonJayamanne/vscodeJupyter)

With rapid continuing growth of Python, Data Science, and Deep Learning communities, Jupyter Notebooks and interactive computing need to be promoted to a first class citizen of VS Code, rather than being an extension developed by a volunteer.

Please refer to Atom + Hydrogen integration, and upcoming JupyerLab projects for ideas on how to integrate Notebooks inside Visual Studio Code.

Thank you!

[MikhailArkhipov]

Related #39492
@DonJayamanne now works for Microsoft BTW :-)

[FuriouslyCurious]

That's awesome @MikhailArkhipov and congratulations @DonJayamanne

We hope to see more amazing REPL, Interactive Computing, and Pythonic things from vsCode now.

[domschl]

See: #3220

[KhurrumMahmood]

Thinking about it a little, I believe the goals should be:

• Have the inline execution work with VS Code's debugger, so we can continue to use the VS dev paradigm.

• Beyond maintaining state in a managed kernel, have options to roll back changes by maintaining a diff of recent actions - because it's not unusual for your code to make a modification that requires you to rerun everything, which wastes time and hurts your flow... And we all want better developer efficiency.

• Jupyter should be supported because of its existing ecosystem and adoption, but beyond that, there should be a focus on facilitating more visualizations and easy building of new visualizations through whatever means offer the greatest / fastest results.

• One of the main benefits of Jupyter Notebooks is shareability - and supporting smooth sharing, and pairing on projects, would be great goal for a new extension.

• Asking existing users of the Jupyter extension for suggestions, and creating a roadmap of incremental releases that offer developers cool, desirable features as soon as possible (and with a continuous feedback loop), would be awesome!

Somewhat unrelated: There are people working on improving Data Science dev workflows by making it possible to use R and Python together in a single IDE. If that were done well, it would be a big win!

Obviously, I just assume that this is 100% worth doing, and MS should dedicate a whole team to this! ;)

[yoDon]

@nojvek also has a repo with VS Code support for converting between .py and .ipynb on the fly

https://github.com/nojvek/vscode-ipynb-py-converter

[jflam]

I had some students at ICL build a prototype of this idea - here's their plugin: https://marketplace.visualstudio.com/items?itemName=neuron.neuron-IPE

Feedback welcome!

[jamescasbon]

I agree that first class notebook support would be massive, and I looked into adding it with existing extension points. However, I don't think its possible.

Motivation: many data scientists use ipython notebook as remote execution environment. However, using the notebook separate to existing ide introduces friction when trying to do the most obvious tasks (eg moving a function prototyped in a notebook into library code). It also means using a separate editor/debugger setup. Currently, jupyter is adding more and more ide features - but first class support for notebooks from an ide would be a lot better.

Suggestion, we need an vs code extension that can:

• render the notebook format [1] https://ipython.org/ipython-doc/3/notebook/nbformat.html showing alternating source input and html output cells.
• execute code on a notebook and update the view [2]

There is support for both of these individually via plugins. With the following drawbacks:

[1] is done via virtual documents or webviews, I believe neither is capable of presenting an editable view of the notebook. I looked into the filesystemprovider extension point. Using this it would be possible to transform the raw json notebook into a reasonable looking editable file with interleaved input and output. It could then use the code lens to present actions for each cell. However, this still doesn't quite do enough: the output cells can contain html (eg inline images) and I don't see a way of rendering inline html in a source document.

[2] executing the code and putting in a virtual document is already working well in other plugins. Therefore, this part should be simple enough given support for [1].

See also #41775

[gramster]

Note that there is an interactive experience in the Python extension now which includes import/export to .ipynb.

https://visualstudiomagazine.com/articles/2019/02/04/vs-code-python.aspx

[xgdgsc]

Importing / exporting feels weired, I don' t want another file.

[stonebig]

I did some experience, and on my win10 pc, Jupyter seems about 2 times slower at first try when used from vscode
... is it possible/normal ?

from vscode

from a jupyter notebokk, the same code

.