nbgather: 🧽✨ Spit shine for Jupyter notebooks

Tools for cleaning code, recovering lost code, and comparing versions of code in Jupyter Lab.

Download the alpha extension with the following command:

jupyter labextension install nbgather

Then you can clean and compare versions of your code like so:

Want to try it out first? Play around with nbgather on an example notebook on BinderHub.

Did the install fail? Make sure Jupyter Lab is up-to-date, and that you are running Jupyter Lab from Python 3.

This project is in alpha: The code this collects will sometimes be more than you want. It saves your a history of all code you've executed and the outputs it produces to the notebook's metadata. The user interface has a few quirks.

Help us make this a real, practical, and really useful tool. We welcome any and all feedback and contributions. We are particularly in need of the opinions and efforts of those with a penchant for hacking code analysis.

Usage Tips

Can it extract more precise slices of code? Yes. First submit a pull request telling us the desired extraction behavior, so we can incorporate this behavior into the tool.

Meanwhile, you can help the backend make more precise slices by telling the tool which functions don't modify their arguments. By default, the tool assumes that functions change all arguments they're called with, and the objects they're called on. To edit the slicing rules, open the Advanced Settings Editor in the Jupyter Lab Settings menu and choose the "Code Gathering Tools" tab. In your user-defined settings, override rules, following this format to specify which functions don't modify their arguments.

How do I clear the notebook's history? Open up your .ipynb file in a text editor, find the history key in the top-level metadata object, and set history to [].

Contributing

To run the development version of nbgather, run:

git clone <this-repository-url>  # clone the repository
jlpm install                     # download dependencies
jupyter labextension link .      # install this package in Jupyter Lab
jlpm run watch                   # automatically recompile source code
jupyter lab --watch              # launch Jupyter Lab, automatically re-load extension

This requires npm version 4 or later, and was tested most recently with Node v9.5.0.

Submit all change as a pull request. Feel free to author the the lead contributor (Andrew Head, andrewhead@berkeley.edu) if you have any questions about getting started with the code or about features or updates you'd like to contribute.

Also, make sure to format the code and test it before submitting a pull request, as described below:

Formatting the code

Before submitting a pull request with changed code, format the code files by running jlpm run format:all.

Testing the code

The tests assume you have Google Chrome installed on your computer. Because this plugin depends on Jupyter Lab and in turn on browser functionality, some of these tests need a browser to run in.

To run the tests from the command line, call:

jlpm run test

Wait a few seconds while the code compiles, and then you should see the results of running the tests. The process will continue to live after the tests finish running---it will recompile and re-run the tests whenever the test code changes. Type Ctrl+C to abort the command at any time.

Note that running tests with this command may interfere with you opening Chrome browsers. If that happens, cancel the command, open Chrome, and then restart the command.

To debug the tests, call:

jlpm run test:debug

This will launch a Chrome window. Click the DEBUG button in the page that opens. Launch the Chrome developer tools (View -> Developer -> Developer Tools). The "Console" will show the test results, with one line for each test. In the "Sources" tab, you can open scripts using the file prompt (Cmd + P on Mac, Ctrl + P on Windows) and set breakpoints in the code. When you refresh the page, the tests will be run again, and the debugger will trigger when the first breakpoint is reached.

Troubleshooting

Here are some tips for dealing with build errors we've encountered while developing code gathering tools:

• Errors about missing semicolons in React types files: upgrade the typescript and ts-node packages
• Conflicting dependencies: upgrade either the Python Jupyter Lab (may require Python upgrade to Python 3 to get the most recent version of Jupyter Lab) or the Jupyter Lab npm pacakges
• Other build issues: we've found some issues can be solved by just deleting your node_modules/ directory and reinstalling it.