A container and a wrapper script around flake8 to validate python code within Jupyter notebooks.
Motivation
An easy way to automate the flake8 code checks over the code blocks defined in a Jupyter notebook.
Example usage
jobs:
flake8:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v2
- uses: mhitza/flake8-jupyter-notebook@v1
Implementation details
There is an existing project, called flake8-nb that performs the same task as this action. While initial implementation tried to wrapped the annotation script around that utility, it was abandoned and instead flake8 was used instead because:
- flake8-nb did not report absolute line number within the notebook file, instead it reported only relative line numbers within the checked code block.
- A notebook might be checked in the repository without the code cells having been evaluated. In that case flake8-nb would report multiple cells without a number, and tracking back from the reported error to absolute line numbers became a more difficult task than wrapping around flake8.
In order to check the notebook, the annotate script keeps track of all the various code blocks within the notebook, concatenates them into a single file and then pipes it as input to flake8.
Known limitations
First of it only supports version format 4 for notebooks. It will just silently skip over other notebook formats, as it's using regular expressions around indentation level to extract source blocks. If you're of any JavaScript JSON parser that keeps track of the source line parsed I'd be happy to hear about it.
Because of implementation details and Jupyter notebook specifics, some warnings and errors reported by flake8 are ignored. The following list is not necessarily exhaustive and prone to be updated based on more testing.
- E302 Expected 2 blank lines when reported for the first line of a code block.
- E305 Expected 2 blank lines after end of function or class, as with E302 just when it's reported for the first line of a code block.
- E402 Module level import not at top of file. In the tested notebooks different code sections will import modules just when needed.