Note: This is a new version of a previously published blog post, updated for the current Android API level (28). Thanks Grayson Parrelli for authoring the original post!
Have you ever accidentally checked on a typo that broke your Android build or unknowingly broke an important use case with a new change? Continuous integration is a way for developers to avoid these headaches, allowing you to confirm that changes to your app compile, and your tests pass before they're merged in.
GitLab CI/CD is a wonderful continuous integration built-in solution, and in this post we'll walk through how to set up a basic config file (.gitlab-ci.yml
) to ensure your Android app compiles and passes unit and functional tests. We assume that you know the process of creating an Android app, can write and run tests locally, and are familiar with the basics of the GitLab UI.
Our sample project
We'll be working with a real-world open source Android project called Materialistic to demonstrate how easy it is to get up and running with GitLab CI for Android. Materialistic currently uses Travis CI with GitHub, but switching over is a breeze. If you haven't seen Materialistic before, it's a fantastic open source Android reader for Hacker News.
Testing
Unit tests are the fundamental tests in your app testing strategy, from which you can verify that the logic of individual units is correct. They are a fantastic way to catch regressions when making changes to your app. They run directly on the Java Virtual Machine (JVM), so you don't need an actual Android device to run them.
If you already have working unit tests, you shouldn't have to make any adjustments to have them work with GitLab CI. Materialistic uses Robolectric for tests, Jacoco for coverage, and also has a linting pass. We'll get all of these easily running in our .gitlab-ci.yml
example except for Jacoco, since that requires a secret token we do not have - though I will show you how to configure that in your own projects.
Setting up GitLab CI
We want to be able to configure our project so that our app is built, and it has the complete suite of tests run upon check-in. To do so, we have to create our GitLab CI configuration file, called .gitlab-ci.yml
, and place it in the root of our project.
So, first things first: If you're just here for a snippet to copy-paste, here is a .gitlab-ci.yml
that will build and test the Materialistic app:
image: openjdk:8-jdk
variables:
ANDROID_COMPILE_SDK: "28"
ANDROID_BUILD_TOOLS: "28.0.2"
ANDROID_SDK_TOOLS: "4333796"
before_script:
- apt-get --quiet update --yes
- apt-get --quiet install --yes wget tar unzip lib32stdc++6 lib32z1
- wget --quiet --output-document=android-sdk.zip https://dl.google.com/android/repository/sdk-tools-linux-${ANDROID_SDK_TOOLS}.zip
- unzip -d android-sdk-linux android-sdk.zip
- echo y | android-sdk-linux/tools/bin/sdkmanager "platforms;android-${ANDROID_COMPILE_SDK}" >/dev/null
- echo y | android-sdk-linux/tools/bin/sdkmanager "platform-tools" >/dev/null
- echo y | android-sdk-linux/tools/bin/sdkmanager "build-tools;${ANDROID_BUILD_TOOLS}" >/dev/null
- export ANDROID_HOME=$PWD/android-sdk-linux
- export PATH=$PATH:$PWD/android-sdk-linux/platform-tools/
- chmod +x ./gradlew
# temporarily disable checking for EPIPE error and use yes to accept all licenses
- set +o pipefail
- yes | android-sdk-linux/tools/bin/sdkmanager --licenses
- set -o pipefail
stages:
- build
- test
lintDebug:
stage: build
script:
- ./gradlew -Pci --console=plain :app:lintDebug -PbuildDir=lint
assembleDebug:
stage: build
script:
- ./gradlew assembleDebug
artifacts:
paths:
- app/build/outputs/
debugTests:
stage: test
script:
- ./gradlew -Pci --console=plain :app:testDebug
Well, that's a lot of code! Let's break it down.
Understanding .gitlab-ci.yml
Defining the Docker Image
image: openjdk:8-jdk
This tells GitLab Runners (the things that are executing our build) what Docker image to use. If you're not familiar with Docker, the TL;DR is that Docker provides a way to create a completely isolated version of a virtual operating system running in its own container. Anything running inside the container thinks it has the whole machine to itself, but in reality there can be many containers running on a single machine. Unlike full virtual machines, Docker containers are super fast to create and destroy, making them great choices for setting up temporary environments for building and testing.
This Docker image (openjdk:8-jdk
) works perfectly for our use case, as it is just a barebones installation of Debian with Java pre-installed. We then run additional commands further down in our config to make our image capable of building Android apps.
Defining variables
variables:
ANDROID_COMPILE_SDK: "28"
ANDROID_BUILD_TOOLS: "28.0.2"
ANDROID_SDK_TOOLS: "4333796"
These are variables we'll use throughout our script. They're named to match the properties you would typically specify in your app's build.gradle
.
ANDROID_COMPILE_SDK
is the version of Android you're compiling with. It should matchcompileSdkVersion
.ANDROID_BUILD_TOOLS
is the version of the Android build tools you are using. It should matchbuildToolsVersion
.ANDROID_SDK_TOOLS
is a little funny. It's what version of the command line tools we're going to download from the official site. So, that number really just comes from the latest version available there.
Installing packages
before_script:
- apt-get --quiet update --yes
- apt-get --quiet install --yes wget tar unzip lib32stdc++6 lib32z1
This starts the block of the commands that will be run before each job in our config.
These commands ensure that our package repository listings are up to date, and it installs packages we'll be using later on, namely: wget
, tar
, unzip
, and some packages that are necessary to allow 64-bit machines to run Android's 32-bit tools.
Installing the Android SDK
- wget --quiet --output-document=android-sdk.zip https://dl.google.com/android/repository/sdk-tools-linux-${ANDROID_SDK_TOOLS}.zip
- unzip -d android-sdk-linux android-sdk.zip
- echo y | android-sdk-linux/tools/bin/sdkmanager "platforms;android-${ANDROID_COMPILE_SDK}" >/dev/null
- echo y | android-sdk-linux/tools/bin/sdkmanager "platform-tools" >/dev/null
- echo y | android-sdk-linux/tools/bin/sdkmanager "build-tools;${ANDROID_BUILD_TOOLS}" >/dev/null
Here we're downloading the Android SDK tools from their official location, using our ANDROID_SDK_TOOLS
variable to specify the version. Afterwards, we're unzipping the tools and running a series of sdkmanager
commands to install the necessary Android SDK packages that will allow our app to build.
Setting up the environment
- export ANDROID_HOME=$PWD/android-sdk-linux
- export PATH=$PATH:$PWD/android-sdk-linux/platform-tools/
- chmod +x ./gradlew
# temporarily disable checking for EPIPE error and use yes to accept all licenses
- set +o pipefail
- yes | android-sdk-linux/tools/bin/sdkmanager --licenses
- set -o pipefail
Finally, we wrap up the before_script
section of our config with a few remaining tasks. First, we set the ANDROID_HOME
environment variable to the SDK location, which is necessary for our app to build. Next, we add the platform tools to our PATH
, allowing us to use the adb
command without specifying its full path, which is important when we run a downloaded script later. Next, we ensure that gradlew
is executable, as sometimes Git will mess up permissions.
The next command yes | android-sdk-linux/tools/bin/sdkmanager --licenses
is responsible for accepting the SDK licenses. Because the unix yes
command results in an EPIPE error once the pipe is broken (when the sdkmanager quits normally), we temporarily wrap the command in +o pipefile
so that it does not terminate script execution when it fails.
Defining the stages
stages:
- build
- test
Here we're defining the different stages of our build. We can call these anything we want. A stage can be thought of as a group of jobs. All of the jobs in the same stage happen in parallel, and all jobs in one stage must be completed before the jobs in the subsequent stage begin. We've defined two stages: build
and test
. They do exactly what you think: the build
stage ensures the app compiles, and the test
stage runs our unit and functional tests.
Building the app
lintDebug:
stage: build
script:
- ./gradlew -Pci --console=plain :app:lintDebug -PbuildDir=lint
assembleDebug:
stage: build
script:
- ./gradlew assembleDebug
artifacts:
paths:
- app/build/outputs/
This defines our first job, called build
. It has two parts - a linter to ensure that the submitted code is up to snuff, and the actual compilation of the code (and configuration of the artifacts
that GitLab should expect to find). These are run in parallel for maximum efficiency.
Running tests
debugTests:
stage: test
script:
- ./gradlew -Pci --console=plain :app:testDebug
This defines a job called debugTests
that runs during the test
stage. Nothing too crazy here about setting something simple like this up!
If we had wanted to get Jacoco also working, that would be very straightforward. Simply adding a section as follows would work - the only additional thing you'd need to do is add a secret variable containing your personal COVERALLS_REPO_TOKEN
:
coverageTests:
stage: test
script:
- ./gradlew -Pci --console=plain jacocoTestReport coveralls
Run your new CI setup
After you've added your new .gitlab-ci.yml
file to the root of your directory, just push your changes to the appropriate branch and off you go! You can see your running builds in the Pipelines tab of your project. You can even watch your build execute live and see the runner's output, allowing you to debug problems easily.
After your build is done, you can retrieve your build artifacts:
- First, click on your completed build, then navigate to the Jobs tab:
From here, simply click on the download button to download your build artifacts.
Conclusion
So, there you have it! You now know how to create a GitLab CI config that will ensure your app:
- Compiles
- Passes tests
- Allows you to access your build artifacts (like your APK) afterwards.
You can take a look at my local copy of the Materialistic repository, with everything up and running, at this link
Enjoy your newfound app stability :)