An example of automatically updating GitHub Pages when you're using Travis CI.
Clone or download
steveklabnik Merge pull request #13 from DavidBruant/patch-1
Changing npm's version of travis-encrypt usage
Latest commit 37f7a7e Jan 9, 2017
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
img First draft of text. Dec 16, 2014
.gitignore Improve gitignore Dec 16, 2014
.travis.yml use real token Dec 16, 2014
Makefile Fix broken image. Dec 16, 2014
README.md Changing npm's version of travis-encrypt usage Jan 9, 2017
deploy.sh Add .travis.yml and deploy script Dec 16, 2014

README.md

Automatically Update Github Pages with Travis

Build Status

Do you want to update Github Pages automatically, and use Travis CI? You've come to the right place.

Both versions:

The problem

Here's a few things, which when combined, cause a problem.

  1. Checking in generated files is considered poor practice.
  2. Often, web pages aren't in HTML directly: they're generated from some other file.
  3. master is the default branch of git repositories.
  4. gh-pages is the default branch for Github Pages.

Our source files end up on one branch, but we need to move the generated files to another branch. And of course, we don't want to just do this on every build, but on successful CI builds of master. Whew!

The solution

Follow these steps.

Ensure you have gh-pages

You want to make sure your branch already exists.

$ git checkout master
$ git checkout --orphan gh-pages
$ git push origin -u gh-pages
$ git checkout master

Easy enough.

Find out your Github API token

Click this link to generate a new Personal access token. You might need to re-enter your password.

You'll need to check some boxes. Check these ones:

github token setting

That's right, just repo. If your repository is public, you can set public_repo instead.

GitHub will create the token, and show you a flash with the value.

THIS IS THE ONLY TIME YOU GET TO SEE THIS SO DON'T CLICK AWAY IMMEDIATELY!

You'll need to copy this token into someplace you trust. I wrote mine down, so I could just light the paper on fire afterward. 😉. It'll never be shown to you after this time, so it's important to double-check your work.

Set up Travis

There are multiple ways to do this.

Set the variables on the Travis-CI dashboard

  • Go to the settings page of your repository on https://travis-ci.org/
  • In the Environment Variables section set a variable with the name of GH_TOKEN and the value of your personal access token.

Set the variables in the .travis.yml file

With Node.js and Python 3.x installed:

$ npm install travis-encrypt -g
$  travis-encrypt -r username/repository GH_TOKEN=[the token you created before]

With Ruby installed:

$ gem install travis
$   travis encrypt -r username/reponame GH_TOKEN=[the token you created before] --add

Note: that I put some spaces before the travis command. If you have bash configured in this common way, this makes sure the command doesn't end up in your Bash History. Can't be too safe with those tokens.

Note: You'll need to have enabled travis for your repo before this.

Check out this page to read more about variable encryption in Travis.

Edit your .travis.yml

Here's what this should look like:

language: something
install:
  - npm install
script:
  - make check
  - make generate
after_success:
  - bash deploy.sh
env:
  global:
    - secure: "oFD/tic8JAwpMXuMDBZXV4ot6w1NLWvHQnrDKmUHSMQJC1cbbrR1p5q8XayfjtmdqQdFQmIfM6YHEKeHw//ypgObWjYS8q00OaaMDXPTdmgr1Ee4nhgkkDihT+kVij0rn96W/QvyAVoaV5hJoyUr3Nhk+mnHEYm3M+Q3LAQglRg="

Let's go over this, line by line:

language: something

This should be set to whatever the language is of your project. What happens if your project's build tool is different than your project itself? You may need to add an install line to install the other tools. Like for example things listed in package.json.

This should look something like this if you use Node.js:

language: node_js
node_js:
  - "4.1"
install:
  - npm install

Next, our actual build:

script:
  - make check
  - make generate

This changes based on whatever your build actually is. I show this section because you will generally want to run more than one command: one to run the tests one to build your project, and one to build the actual documentation.

after_success:
  - bash deploy.sh

Ok so now we have a successful build so we want to start the deploy. Note that we can have a different approach:

after_success:
  - test $TRAVIS_PULL_REQUEST == "false" && test $TRAVIS_BRANCH == "master" && bash deploy.sh

Here we want to check out where we are. We only want to update Github Pages if we're building the master branch of the original repository, so we have to check $TRAVIS_PULL_REQUEST and $TRAVIS_BRANCH.

If we are here, we run bash deploy.sh. What's the contents of deploy.sh? We'll talk about that in a moment. We have one more line to cover:

env:
  global:
    - secure: "oFD/tic8JAwpMXuMDBZXV4ot6w1NLWvHQnrDKmUHSMQJC1cbbrR1p5q8XayfjtmdqQdFQmIfM6YHEKeHw//ypgObWjYS8q00OaaMDXPTdmgr1Ee4nhgkkDihT+kVij0rn96W/QvyAVoaV5hJoyUr3Nhk+mnHEYm3M+Q3LAQglRg="

This, of course, should use the value from travis encrypt from before. Remember how we encrypted GH_TOKEN=... before? This will ensure that our GH_TOKEN variable is set to the unencrypted value. That sounds scary, but Travis will not set this on forks or pull requests, so that someone can't just submit a PR that echoes the value out.

Set up deploy script

Okay, next, we need to add a deploy.sh to our repository. You'll need to tweak this slightly for your setup, but here's the basic idea:

#!/bin/bash

set -o errexit -o nounset

if [ "$TRAVIS_BRANCH" != "master" ]
then
  echo "This commit was made against the $TRAVIS_BRANCH and not the master! No deploy!"
  exit 0
fi

rev=$(git rev-parse --short HEAD)

cd stage/_book

git init
git config user.name "Steve Klabnik"
git config user.email "steve@steveklabnik.com"

git remote add upstream "https://$GH_TOKEN@github.com/rust-lang/rust-by-example.git"
git fetch upstream
git reset upstream/gh-pages

echo "rustbyexample.com" > CNAME

touch .

git add -A .
git commit -m "rebuild pages at ${rev}"
git push -q upstream HEAD:gh-pages

Let's do it, paragraph by paragraph:

#!/bin/bash

The standard shebang line. We don't really need to set this, as we execute it with bash deploy.sh, but I like to put it in anyway.

set -o errexit -o nounset

This sets two options for the shell to make the script more reliable:

  • errexit: stop executing if any errors occur, by default bash will just continue past any errors to run the next command
  • nounset: stop executing if an unset variable is encountered, by default bash will use an empty string for the value of such variables.
if [ "$TRAVIS_BRANCH" != "master" ]
then
  echo "This commit was made against the $TRAVIS_BRANCH and not the master! No deploy!"
  exit 0
fi

Here we ensure that we only deploy when we commit against the master branch, if not we just simply abort the deploy, no errors. So this way we can see the result of the tests when we make pull request between different branches or commit against a different branch than the master.

Note: This only works if in .travis.yml we used the unconditional deploy.

after_success:
  - bash deploy.sh
rev=$(git rev-parse --short HEAD)

This sets a variable, rev, with the short hash of HEAD. We'll use this later in a commit message.

cd _book

We need to cd into wherever our website built. With Jekyll, it's _site. But do whatever.

git init
git config user.name "Steve Klabnik"
git config user.email "steve@steveklabnik.com"

First, we initialize a new git repository. Yes, a new one. You'll see.

We then set our user name and user email. This person will have done the commits that go to gh-pages. It's not a default branch, so don't worry, GitHub doesn't count these commits as contributions for your graph.

git remote add upstream "https://$GH_TOKEN@github.com/me/project.git"
git fetch upstream
git reset upstream/gh-pages

Next, we add a remote, named upstream, and we set it to our project. But we also interpolate that $GH_TOKEN variable, which will allow us to push to this repository later.

We then fetch it and reset to the gh-pages branch. Now, git sees this new repository as just some files that change your upstream gh-pages branch.

echo "myproject.com" > CNAME

Sometimes, you'll need some extra files. A CNAME is common, which sets a custom domain up. You'll need to run whatever commands generate those files for you.

touch .

We then touch everything, so that git considers all of our local copies fresh.

git add -A .
git commit -m "rebuild pages at ${rev}"
git push -q upstream HEAD:gh-pages

We then add all changes, commit them, using our rev from earlier, and then push to upstream. The -q keeps this a bit more quiet, and you can control the noisiness of all these different git commands with a judicious sprinkling of -q.

Success!

That's it! Commit this all, and push. Travis should now do its magic, and everything will update!

One Drawback

One drawback of this is that if you have a build matrix that builds your project with multiple versions of your platform, you'll end up with the same number of pages builds. Which seems redundant.

I think I could take advantage of Travis' "deploy" feature to fix this, but I'm not sure how.

Feedback please

I'd love to know if there's a better way to do any of this. In particular, I'd love to add the local git repo rather than the one from GitHub when fetching the upstream, but since Travis checks out a bare repository, it doesn't seem possible. Please open an issue or PR to show me how to do it better!