Use Jupytext notebooks

The idea is to store notebooks as markdown files, so that under Git it's easier to understand the changes.

It's the very first time I'm using Jupytext, and I don't guarantee I did all correctly:

• I converted all notebooks with jupytext --to md, and then removed the original ipynb files.
• I added Juytext in the conda environment.
• I added a jupytext-configfile to ensure the new Markdown files are directly interpreted as notebooks (I don't use the "pairing" possibility that keeps a ipynb file along the Markdown file

Feel free to tell me if it can be done in a better way!

A known limitation in current configuration is that with Binder the Markdown files are not loaded by default as notebooks (but with 'open with' it works...)

added Jupyter conda labels

requested review from @vrouvrea

assigned to @sgilles

I tried, seems to work fine, except I have to right click on the markdown files, and Open With / Notebook for the file to be correctly open. Otherwise I only see the markdown file as plain text.

Strange thing is the icon is correct in the File Browser left panel view...

Locally I manage to open notebook directly; I should though update the doc.

I'll do this after the holidays!

added 1 commit

• 9132efe2 - Fix the configuration file for Jupytext so that Markdown notebooks are...

Compare with previous version

It was indeed not running properly if I tried in a new environment.

I have therefore patched the Jupytext configuration file: both its name and the syntax used were incorrect.

Seems to work properly for me locally; can't manage to run Binder properly for some reason (404 error obtained??? I'll investigate this)

@vrouvrea Please nonetheless tell me if it's the same for you locally!

UPDATE On subsequent try I managed to create and run a Binder instance; the files were properly rendered as notebooks but it didn't manage to connect the kernel... Doesn't seem to be a good day for Binder?

Still not ok for me. We should take some time to clearly identify the issue.

Ok that's weird as it works well on @poli 's former Linux laptop. But on the other hand I really don't manage to make it run properly on Binder (but Binder is capricious - @poli and I had to use an ugly workaround to keep finding properly the kernels as conda installation doesn't seem to work properly).

As you say the best is to take a bit of time to look it together next time we're both at Turing

added 1 commit

• 8099c841 - Try changing ordering in environment (in Binder JupyterLab doesn't seem to...

Compare with previous version

I have checked this morning:

• Seems to work as expected on Linux as well

• For Binder, changing the ordering of requirements in the yaml file seems to improve launch of notebook (for some reason if JupyterLab was before Jupytext the Lab version was not generated and the url with Lab gave a 404). However in the Binder instance the toml file doesn't seem to be considered, and the files are not loaded as notebooks.

Weirdly enough, shuffling the dependencies in the Yaml may induce lack of kernels found in Binder

@poli suggested a workaround for this:

• Create a binder/postBuild file in the repository
• Inside explicitly call conda to install xeus-cling:

#!/bin/bash

conda install -y xeus-cling -c conda-forge

added 4 commits

• 150b1899 - Try replacing jupytext.toml by pyproject.toml.
• e8f4eacd - Add postBuild file for Binder to ensure all is correctly installed.
• 88be5a9c - Try adding toml symbolic link in binder folder.
• 4399c2fb - Try labconfig hack used for Binder by Jupytext for its own Binder image.

Compare with previous version

Following @poli's suggestion, I copied the hack used by Jupytext for their own Binder instance and it seems to work

added 7 commits

• 556f3ee1 - Try adding Binder requirement file.
• 3c8461fd - Add version for Jupytext in requirements.
• f3267b56 - Try renaming binder directory .binder.
• ad43c25d - Moving Xeus-cling in environment.yml.
• ba1d915a - #113 (closed) Rename .binder binder.
• 52f338b8 - #113 (closed) Remove activate command, and add -y to conda install.
• 9c809fda - #113 (closed) Try using pyproject.toml.

Compare with previous version

Although it would be much nicer to get proper diffs in Gitlab interface for notebooks, make it work by default in all environments (macOS, Linux and Binder alike) is really a pain. Let's drop it for the time being.

closed