Jupytext's CM uses parent CM's get and save methods to read text files

[mwouts]

With this PR we stop using mock to replace the nbformat.reads and writes functions with Jupytext's ones. Instead we use the parent contents manager to load/write the text files and parse them with Jupytext.

As a side effect, we don't use any more private methods from the FileContentsManager or LargeFileManager, and Jupytext can work on top of other contents managers like Jupyter-FS.

This is a follow-up on #621 and will hopefully close #618

[codecov]

Codecov Report

Merging #634 into master will increase coverage by 0.00%.
The diff coverage is 98.91%.

@@           Coverage Diff           @@
##           master     #634   +/-   ##
=======================================
  Coverage   99.18%   99.18%           
=======================================
  Files          94       94           
  Lines        9307     9359   +52     
=======================================
+ Hits         9231     9283   +52     
  Misses         76       76           

Impacted Files	Coverage Δ
tests/test_trust_notebook.py	98.98% <94.73%> (-1.02%)	⬇️
jupytext/contentsmanager.py	98.51% <98.07%> (+0.06%)	⬆️
tests/test_cli.py	100.00% <100.00%> (ø)
tests/test_cm_config.py	100.00% <100.00%> (ø)
tests/test_contentsmanager.py	100.00% <100.00%> (ø)
tests/test_escape_magics.py	100.00% <100.00%> (ø)
tests/test_read_simple_percent.py	100.00% <100.00%> (ø)
tests/test_save_multiple.py	100.00% <100.00%> (ø)
tests/utils.py	100.00% <100.00%> (ø)
jupytext/config.py	97.51% <0.00%> (+0.62%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update aea84d9...5a21182. Read the comment docs.

[mwouts]

With this version I see a warning: 'dict' object has no attribute 'metadata' when I save paired notebooks.

[telamonian]

@mwouts I'll take a look

[mwouts]

Thank you @telamonian , well sure it is a great idea if you want to have a look at this. With this PR the contents manager will be much cleaner - the implementation in Jupytext <= 1.6.0 works well in practice but is a bit hacky!

The tests on this PR pass but as mentioned above the contents manager does not work in the real world. I am a bit surprised that the tests don't catch this... I'll see tonight if I can reproduce the problem in the test framework, and report here.

[telamonian]

@mwouts I tracked the problem down and fixed it. It actually seems like jupyter-fs and jupytext are working together well now. All the details are at #635.

I'm trying to put together some tutorials for how to debug Jupyter-related code. So as an experiment/rough draft I filmed myself working on this bug and uploaded it to youtube.

• part 1: https://www.youtube.com/watch?v=VB-0o6JZVE4
  • tutorial and basic setup
• part 2: https://www.youtube.com/watch?v=asZukV-JEso
  • applying the stuff form part 1 to solve the bug from Jupytext's CM uses parent CM's get and save methods to read text files #634 (comment), beginning to end

If you have a sec, check it out

[joequant]

Great job!!!!

Is there any place we can start a wiki to talk about debugging and testing techniques?

The tools that I use are Firefox developer and KernelSpy, and I find that a lot of bugs can be tracked down by just running these two programs and seeing where you have error messages, or weird behaviour.

Also, I have some hints as to guessing where the bugs often are.

[mwouts]

Hi Max, just wanted to thank you for spending time on this project. I've watched the two videos, they are very interesting! Nice way to spend time together 😄

A few quick comments:

• I'm sure I will reuse your settings from the first one next time I have to debug the TypeScript extension 😄 Currently for Python I am very happy with PyCharm, but for the JavaScript / TypeScript part I was missing a better development environment than the browser...
• Seeing you scroll among the code and the latest commits is a great incentive to write clearer code and commits !
• Oh yes, using mock was a hack, sorry about this... I was just learning Python (and pytest, hence mock...) at that time.... and it worked too well - I mean it took two years before a real limitation appeared... But of course I'm glad to have this better implementation now.
• Just like you, I love interactive debugging, but maybe I am even more fond of the tests - In my experience they help ensure that I won't need to debug another time. What do you think?

Looking forwards to listening to your talk at the next JupyterCon 👋