Test jupytext against a large collection of notebooks

[mwouts]

It would be great to test jupytext's round trip conversion on a large collection of notebooks. Take for instance Jake Vanderplas' Python for Data Science Handbook.

[mwouts]

That was an interesting exercise! On the way I encountered a YAML RepresenterError for one of the notebooks (fixed with f1285db). I also had to improve significantly the reporting on the round trip conversion.

Now the results:

• Round trip conversion is consistent with Jupytext's expectations. Each of the below returns with no error:

jupytext --test *.ipynb --to py:light
jupytext --test *.ipynb --to py:percent
jupytext --test *.ipynb --to py:sphinx
jupytext --test *.ipynb --to md
jupytext --test *.ipynb --to Rmd

• Round trip conversion is not strictly the identity. Conversion to Python scripts removes one final blank line in cell 9 of 03.06-Concat-And-Append.ipynb. And cell metadata: 'collapsed', 'scrolled' are absent from the script. Conversion to Markdown or R Markdown removes cell metadata on markdown cells. Thus cell metadata 'collapsed', 'deletable', 'editable' are missing. And markdown cell 35 of notebook 01.07-Timing-and-Profiling.ipynb is splitted into multiple cells (as the cell contains two consecutive blank lines)

jupytext --test-strict *.ipynb --to py:light # missing cell metadata, one final blank line removed in one code cell
jupytext --test-strict *.ipynb --to md       # missing cell metadata, one markdown cell splitted

• Round trip conversion of paired notebook is closer to identity - cell metadata not represented in the scripts is taken from the ipynb file

jupytext --test-strict *.ipynb --to py:light # one final blank line removed in cell 9 of 03.06-Concat-And-Append.ipynb
jupytext --test-strict *.ipynb --to md       # last markdown cell splitted in 01.07-Timing-and-Profiling.ipynb

I also have documented Jupytext's definition of unchanged notebook for the round trip conversion, and the --test, --test-strict and -x arguments in the README.