The issue

Documentation in the software world can be pretty bad. It’s all too common to find a promising git project with several hundred stars and nothing but a README to document package use. While tools like sphinx have made auto-documentation more straight-forward for developers, hosting said documentation poses another hurdle.

When it comes to publishing & hosting documentation, Github solves the problem with its gh-pages functionality, which allows you to host static web pages from a branch. But if you’ve ever used it, you’re probably aware of the headaches it can cause.

In this post, I’m going to share a workflow I’ve developed to automate my documentation builds and deployments to gh-pages using Circle CI.

The first push to gh-pages:

Before we automate the builds, it’s helpful to have the gh-pages branch on the server. Make sure you have all of your changes committed and pushed before proceeding.

From the master (or your default) branch, we’re going to checkout a new orphaned branch called gh-pages. This is a magic branch name, so it’s important you call it this, otherwise you’ll have to do some voodoo in your project settings to make this all work.

$ git checkout --orphan gh-pages

Now, we’re going to delete everything. Everything except the .git directory!

This sounds scary, but if you’ve committed everything, it’s going to be just fine! Here’s a quick hack that will delete all files except the .git directory:

$ find . ! -name ".git/*" -type f -exec rm -f {} +

But depending on whatever else you have in the project folder, you may have several other artifacts or directories to clean up. Just make sure the .git folder goes untouched.

The first push is easy. It’s going to be an empty commit:

$ git add --all
$ git commit --allow-empty
$ git push origin gh-pages

Automating all future doc deployments

Now that the branch exists on remote, we can streamline things. If you don’t already have a Circle account, head on over to the sign-up page and click “Sign Up with Github”. From there, you can setup your .circleci/config.yml add your project to a Circle workflow to build/test on Github pushes. If you’re not familiar with how to do this, follow this tutorial on getting setup with Circle.

Creating a build script for your documentation

Here’s what my build script looks like. It’s located inside the build_tools/circle directory. I’ve littered comments throughout to try to explain each step:


# Fail out on an error
set -e

# This is a hack, but we have to make sure we're only ever running this
# from the top level of the package and not in the subdirectory...
# Shouldn't ever be an issue the way we've got this setup, and you'll
# want to change it a bit to make it work with your project structure.
if [[ ! -f gh_doc_automation/ ]]; then
    echo "This must be run from the gh_doc_automation project directory"
    exit 3

# CD into docs, make them. If you're not using Sphinx, you'll probably
# have a different build script.
cd doc
make clean html EXAMPLES_PATTERN=ex_*
cd ..

# Move the docs to the top-level directory, stash for checkout
mv doc/_build/html ./

# The html/ directory will stay there when we stash
git stash

# Checkout our gh-pages branch, remove everything but .git
git checkout gh-pages
git pull origin gh-pages

# Make sure to set the credentials! You'll need these environment vars
# set in the "Environment Variables" section in Circle CI
git config --global "$GH_EMAIL" > /dev/null 2>&1
git config --global "$GH_NAME" > /dev/null 2>&1

# Remove all files that are not in the .git dir
find . -not -name ".git/*" -type f -maxdepth 1 -delete

# Remove the remaining artifacts. Some of these are artifacts of the
# LAST gh-pages build, and others are remnants of the package itself.
# You will have to amend this to be more specific to your project.
declare -a leftover=(".cache/"

# Check for each left over file/dir and remove it, or echo that it's
# not there.
for left in "${leftover[@]}"
    rm -r "$left" || echo "$left does not exist; will not remove"

# We need this empty file for git not to try to build a jekyll project.
# If your project *is* Jekyll, then this doesn't apply to you...
touch .nojekyll
mv html/* ./
rm -r html/

# Add everything, get ready for commit. But only do it if we're on
# master. If you want to deploy on different branches, you can change
# this.
if [[ "$CIRCLE_BRANCH" =~ ^master$|^[0-9]+\.[0-9]+\.X$ ]]; then
    git add --all
    git commit -m "[ci skip] publishing updated documentation..."

    # We have to re-add the origin with the GH_TOKEN credentials. You
    # will need this SSH key in your environment variables on Circle.
    # Make sure you change the <project>.git pattern at the end!
    git remote rm origin
    git remote add origin https://"$GH_NAME":"$GH_TOKEN""$GH_NAME"/gh_automation.git

    # NOW we should be able to push it
    git push origin gh-pages
    echo "Not on master, so won't push doc"

I’m using Sphinx, so my documentation is built with the doc/Makefile that’s created from the sphinx-quickstart. If you’re using a different framework, you’ll have to make some corrections to this line:

cd doc
make clean html EXAMPLES_PATTERN=ex_*
cd ..

There are several points in the script where you’ll have to make amendments with your git project name to make sure you’re pointing to the right place. You also need the script to be executable before you even push, or you’ll get a build failure!

$ chmod +x build_tools/circle/

Setting environment variables

In the script above, you’ll notice that there are some environment variables you’ll need to have stashed in the “Environment Variables” section of Circle. Otherwise, you’ll get an error like this:

Circle failure

There are three environment variables you’ll need:

  • GH_NAME - Your github username. Mine, for instance, is tgsmith61591
  • GH_EMAIL - The email address you have associated with your github account.
  • GH_TOKEN - This is an SSH key you’re going to have to create on Github. Follow these instructions

All of these are going to be necessary for you to be able to push to your remote from Circle, and the token in particular should be closely guarded.

Amending your config.yml

At this point, you should have a Circle config file setup for your project, and you should have your build script setup. Now we’re going to amend your config with the documentation build/push workflow.

All you need to do is to create a new section in your YAML file. Here’s what mine looks like in the example project:

  - image: python:3.6
  working_directory: ~/gh_doc_automation
  - checkout
  - run: make doc-requirements
  - run: make install
  - run: ./build_tools/circle/

You can use a different Python image if you want, and you’ll want to set the working directory to your project name. Notice also that I’m building from a Makefile. All the Makefile does is alias away a series of commands behind one “target”. Here’s what happens when make doc-requirements is called:

$ python -m pip install -r build_tools/doc/doc_requirements.txt

Using a Makefile is completely a taste thing and is by no means required for this to work. Without a Makefile, my config would look like:

  - image: python:3.6
  working_directory: ~/gh_doc_automation
  - checkout
  - run: python -m pip install -r build_tools/doc/doc_requirements.txt
  - run: python -m pip install -r requirements.txt
  - run: python install
  - run: ./build_tools/circle/

Once you have added the job to your config file, you need to make sure it runs. Add the job to your existing workflow:

  version: 2
      - some-other-job
      - deploy-doc

And that’s all! Next time you push to master, your documentation will be built and deployed! You can view it at:


Final thoughts

This is a simple approach to automating your documentation pushes. Of course, there will be corner cases where this isn’t exactly what you’re after, but it should provide you with what you need to adapt the solution to your requirements. You may also be interested in checking out the way I build versioned documentation in my project, pmdarima.