A container and a wrapper script around flake8 to validate python code within Jupyter notebooks. flake8 will pick up configuration files in your project, but some options are not supported and will cause the action to fail (sometimes silently).
An easy way to automate flake8 code checks over code blocks defined in a Jupyter notebook.
jobs:
flake8:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- uses: mhitza/flake8-jupyter-notebook@v1
with:
debug: 'false' # set 'true' for additional logging
# paths and files to ignore, one regexp rule per line
ignore: |
tests/
test\.ipynb$
There is an existing project, called flake8-nb that performs the same task as this action. While initial implementation tried to wrapp the annotation script around that utility, it was abandoned and instead flake8 was used because:
- flake8-nb did not report absolute line number within the notebook file, instead it reported only relative line numbers within the checked code blocks.
- A notebook might be checked in the repository without the code cells evaluated. In that case flake8-nb would report on 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 source and pipes it into flake8.
Supports version 4 compatible notebook formats. It will just silently skip over other notebook formats, as it's using regular expressions based on indentation level to extract source blocks. If you're of aware of any JavaScript JSON parser that keeps track of the source lines parsed I'd be happy to hear about it.
Due to implementation details and Jupyter notebook specific idiosyncrasies, some warnings and errors reported by flake8 are ignored by default (hardcoded in source code). The following list is not necessarily exhaustive and might change based on testing and issues raised.
- 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, when reported for the first line of a code block.
- E402 Module level import not at top of file. In tested notebooks code sections will tend to import modules just before usage.
- F821 Undefined name name. When the undefined name stands for the Jupyter builtin function
display
The following options which can be defined in a flake8 configuration file are not supported.
Any option that changes the output of flake8: --quiet, --count, --format (only default supported),
--show-source, --statistics.
Anything that relies on filename/paths, as code is passed in to flake8 via stdin.
Thus the following options will have no effect: --exclude, --extend-exclude, --filename, --per-file-ignores
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent