Interface ReadTheDocs and GitHub Actions

I like to use ReadTheDocs to build (and version!) my docs, but I also like to use Jupyter notebooks to write tutorials. Unfortunately, this has always meant that I needed to check executed notebooks (often with large images) into my git repository, causing huge amounts of bloat. Futhermore, the executed notebooks would often get out of sync with the development of the code. No more!!

This library avoids these issues by executing code on GitHub Actions, uploading build artifacts (in this case, executed Jupter notebooks), and then (only then!) triggering a ReadTheDocs build that can download the executed notebooks.

There is still some work required to set up this workflow, but this library has three pieces that make it a bit easier:

1. A GitHub action that can be used to trigger a build for the current branch on ReadTheDocs.
2. A Sphinx extension that interfaces with the GitHub API to download the artifact produced for the target commit hash.
3. Some documentation that shows you how to set all this up!

Usage

The following gives the detailed steps of the process of setting up a project using this workflow. But you can also see a fully functional example in this repository. The documentation source is the docs directory and the .github/workflows directory includes a workflow that is executed to build the docs using this package. The rendered page is available at rtds-action.readthedocs.io.

1. Set up ReadTheDocs

1. First, you'll need to import your project as usual. If you've already done that, don't worry: this will also work with existing ReadTheDocs projects.
2. Next, go to the admin page for your project on ReadTheDocs, click on Integrations (the URL is something like https://readthedocs.org/dashboard/YOUR_PROJECT_NAME/integrations/).
3. Remove existing integrations.
4. Click Add integration and select Generic API incoming webhook.
5. Take note of the webhook URL and token on this page for use later.

2. Set up GitHub Actions workflow

In this example, we'll assume that we have tutorials written as Jupyter notebooks, saved as Python scripts using Jupytext (because that's probably what you should be doing anyways!) in a directory called docs/tutorials.

First, you'll need to add the ReadTheDocs webhook URL and token that you recorded above as "secrets" for your GitHub project by going to the URL https://github.com/USERNAME/REPONAME/settings/secrets. I'll call them RTDS_WEBHOOK_URL (include the https!) and RTDS_WEBHOOK_TOKEN respectively.

For this use case, we can create the workflow .github/workflows/docs.yml as follows:

name: Docs
on: [push, release]

jobs:
  notebooks:
    name: "Build the notebooks for the docs"
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v2

      - name: Set up Python
        uses: actions/setup-python@v2
        with:
          python-version: 3.8

      - name: Install dependencies
        run: |
          python -m pip install -U pip
          python -m pip install -r .github/workflows/requirements.txt

      - name: Execute the notebooks
        run: |
          jupytext --to ipynb --execute docs/tutorials/*.py

      - uses: actions/upload-artifact@v2
        with:
          name: notebooks-for-${{ github.sha }}
          path: docs/tutorials

      - name: Trigger RTDs build
        uses: dfm/rtds-action@v1
        with:
          webhook_url: ${{ secrets.RTDS_WEBHOOK_URL }}
          webhook_token: ${{ secrets.RTDS_WEBHOOK_TOKEN }}
          commit_ref: ${{ github.ref }}

Here, we're also assuming that we've added a pip requirements file at .github/workflows/requirements.txt with the dependencies required to execute the notebooks. Also note that in the upload-artifact step we give our artifact that depends on the hash of the current commit. This is crucial! We also need to take note of the notebooks-for- prefix because we'll use that later.

It's worth emphasizing here that the only "special" steps in this workflow are the last two. You can do whatever you want to generate your artifact in the previous steps (for example, you could use conda instead of pip) because this workflow is not picky about how you get there!

3. Set up Sphinx

Finally, you can edit the conf.py for your Sphinx documentation to add support for fetching the artifact produced by your action. Here is a minimal example:

import os

extensions = [... "rtds_action"]

# The name of your GitHub repository
rtds_action_github_repo = "USERNAME/REPONAME"

# The path where the artifact should be extracted
# Note: this is relative to the conf.py file!
rtds_action_path = "tutorials"

# The "prefix" used in the `upload-artifact` step of the action
rtds_action_artifact_prefix = "notebooks-for-"

# A GitHub personal access token is required, more info below
rtds_action_github_token = os.environ["GITHUB_TOKEN"]

Where we have added the custom extension and set the required configuration parameters.

You'll need to provide ReadTheDocs with a GitHub personal access token (it only needs the public_repo scope if your repo is public). You can generate a new token by going to your GitHub settings page. Then, save it as an environment variable (called GITHUB_TOKEN in this case) on ReadTheDocs.

Development

For now, just a note: if you edit src/js/index.js, you must run npm run package to generate the compiled action source.