Running isort on Jupytext notebooks represented as scripts break some cells

[mwouts]

This is related to #446 - It would be great to be able to run isort easily on a Jupytext notebook, however ATM it does not work that well.

Indeed, isort seems to move the comments preceding an import together with the import. Thus, when we run isort on a .py notebook, imports at the top of the cell are moved along the preceding cell marker (# %% or # +), breaking the corresponding cell.

Possibly one way to fix this is to insert one blank line between the cell marker and the cell content. I'll see if we can add an option to do that, and if isort works better, then. Does this option work well with black? If so, it should be made the default for when a notebook is piped into an other command.

[mwouts]

@timothycrosley implemented two new options in isort, see PyCQA/isort#1338.
With those options I have successfully reorder the imports in a notebook with

jupytext --pipe 'isort - --treat-comment-as-code "# %%" --float-to-top' notebook.ipynb

See this commit: a73f386
I will now test this at a larger scale.

[rijobro]

This still seems to be a problem for me (but only on 1 of many notebooks, not sure what makes it different).

Command was jupytext --pipe 'isort - --treat-comment-as-code "# %%" --float-to-top' notebook.ipynb

I'm using jupytext 1.9.1 and isort 5.7.0.

Before:

After:

[mwouts]

Hi @rijobro , thank you for reporting this. Could you share the py:percent representation of (that part) of the notebook? Especially, is there anything else than # %% on the import cell, e.g. some cell metadata?

[rijobro]

Hi, yes it seems that the cell containing the imports had the following metadata:

   "metadata": {
    "tags": []
   },

And changing it to the following solved the problem

   "metadata": {},