Add Jupytext requirement for binder.

[Carreau]

I think binder looks at env.yaml not at requirements.txt maybe that's the reason that binder don't have jupytext installed

[melissawm]

Let's try - I think it makes sense. Thanks @Carreau !

[melissawm]

Looks like this works - in any case we still need to "Open as notebook"

[bsipocz]

what about jupyterlab-myst, would that fix this?

[rowanc1]

You will still need jupytext, and to open as notebook with jupyterlab-myst, the tools are designed to work together. If you want the experience for users to be simpler, you could potentially render out the (pre-executed) notebooks, and have that be the Jupyter experience that gets started.

[bsipocz]

you could potentially render out

you mean an ipynb notebook? (That is what I do with astroML and really don't like the process/setup to have both mb and ipynb (but also, I need the outputs at least for a couple of notebooks as some are too expensive to execute each time in CI)

[rowanc1]

Yes, that is what I meant is automatically creating those for when people are interacting with the binder instance. I am not suggesting keeping those in git though! That would require some improvements/modifications to CI to enable that though.

[melissawm]

We are doing something similar here: https://github.com/scipy/scipy/blob/d304e95751d52dd327e10634b06c17f0440ba5bc/doc/Makefile#L116

[bsipocz]

We do that here, too, they are then pushed to the gh-pages branch along with the rendered htmls. Could binder be forced to pick up those ones?

https://github.com/numpy/numpy-tutorials/tree/gh-pages/_sources/content

[rossbar]

I'd rather pursue client-side interactivity (ideally on the page itself) than put effort into adding .ipynb-formatted files for binder/jupyterhub support. I'd like to push towards a future where .md-formatted notebooks are natively supported. Jupytext is fantastic, and the implicit behind-the-scenes support for md->ipynb that makes this work currently is a critical component in a lot of the sustainable docs workflows (not only for server-side interactivity, but also for testing/linting, etc.).

[melissawm]

I agree, I think that would be ideal. But as a compromise while that doesn't happen we can have binder point to the ipynb files. I'll try to poke around.

[rowanc1]

In the new mystmd work we have put a lot of work into thebe which enables client-side interactivity (powered by jupyter, binder or jupyterlite). There are some docs on that here:

https://mystmd.org/guide/integrating-jupyter

Because the numpy tutorials are using well-supported packages, I think that adding jupyter: lite to the myst project config, might just work..?

We currently haven't enabled editing directly on the page yet, but it is on our roadmap.

• Thebe integration: Modify sources jupyter-book/mystmd#443

Curious if @rossbar and @melissawm this is getting close to the direction you imagine?

[rossbar]

Curious if @rossbar and @melissawm this is getting close to the direction you imagine?

Indeed - this was one of the features from that was most compelling from my perspective! It's great to see that on-page interactivity is on the roadmap!

[melissawm]

Yes, 100%! Should we follow that issue for updates?

[rowanc1]

I am tracking the request from MyST's side here:

• Tracking: Numpy Tutorials jupyter-book/mystmd#1270