As reported at #715 , the new cell ids introduced in notebooks in nbformat version 4.5 causes trouble with Jupytext. Indeed, the text representation does not store the cell id, and in that case nbformat regenerate a (random) cell id, contributing random differences on text <-> ipynb round trips.
I see at least two possible approaches to address this issue:
- Patch
nbformat to generate non-random cell ids
- Store the cell id in the Jupytext representation of the notebook
Anyone is welcome to contribute ideas or comments!