Google Bazel is needed to compile the code, and an SQL database to house the review metadata. H2 is recommended for development databases, as it requires no external server process.

Getting the Source

Create a new client workspace:

  git clone --recursive
  cd gerrit

The --recursive option is needed on git clone to ensure that the core plugins, which are included as git submodules, are also cloned.


Please refer to Building with Bazel.

Switching between branches

When switching between branches with git checkout, be aware that submodule revisions are not altered. This may result in the wrong plugin revisions being present, unneeded plugins being present, or expected plugins being missing.

After switching branches, make sure the submodules are at the correct revisions for the new branch with the commands:

  git submodule update
  git clean -fdx
If you decide to store your Eclipse/IntelliJ project files in the Gerrit source directories, executing git clean -fdx will remove them and hence screw up your project.

Configuring Eclipse

To use the Eclipse IDE for development, please see Eclipse Setup.

For details on how to configure the Eclipse workspace with Bazel, refer to: Eclipse integration with Bazel.

Configuring IntelliJ IDEA

Please refer to IntelliJ Setup for detailed instructions.

Mac OS X

On Mac OS X ensure "Java For Mac OS X 10.5 Update 4" (or later) has been installed, and that JAVA_HOME is set to the required Java version.

Java installations can typically be found in "/System/Library/Frameworks/JavaVM.framework/Versions".

You can check the installed Java version by running java -version in the terminal.

Site Initialization

After compiling (above), run Gerrit’s 'init' command to create a testing site for development use:

  $(bazel info output_base)/external/local_jdk/bin/java \
     -jar bazel-bin/gerrit.war init -d ../gerrit_testsite
You must use the same Java version that Bazel used for the build. This Java version is available at $(bazel info output_base)/external/local_jdk/bin/java.

During initialization, make two changes to the default settings:

  • Change the listen addresses from '*' to 'localhost' to prevent outside connections from contacting the development instance; and

  • Change the auth type from 'OPENID' to 'DEVELOPMENT_BECOME_ANY_ACCOUNT' to allow yourself to create and act as arbitrary test accounts on your development instance.

Continue through init until it completes. The daemon will automatically start in the background and a web browser will launch to the start page. From here you can sign in as the account created during init, register additional accounts, create projects, and more.

When you want to shut down the daemon, simply run:

  ../gerrit_testsite/bin/ stop

Working with the Local Server

If you need to create additional accounts on your development instance, click 'become' in the upper right corner, select 'Switch User', and then register a new account.

Use the ssh protocol to clone from and push to the local server. For example, to clone a repository that you’ve created through the admin interface, run:

git clone ssh://username@localhost:29418/projectname

Then you’ll be able to create changes the same way users do, with

git push origin HEAD:refs/for/master


Running the Acceptance Tests

Gerrit has a set of integration tests that test the Gerrit daemon via REST, SSH and the git protocol.

A new review site is created for each test and the Gerrit daemon is started on that site. When the test has finished the Gerrit daemon is shutdown.

For instructions on running the integration tests with Bazel, please refer to: Running Unit Tests with Bazel.

Running the Daemon

The daemon can be directly launched from the build area, without copying to the test site:

  $(bazel info output_base)/external/local_jdk/bin/java \
     -jar bazel-bin/gerrit.war daemon -d ../gerrit_testsite \
Please refer to this explanation for details why using java -jar isn’t sufficient.

If you want to debug the Gerrit server of this test site, you can open a debug port (for example port 5005) by inserting


directly after -jar of the previous command. Please refer to Debugging a remote Gerrit server for instructions of how to attach IntelliJ.

Running the Daemon with Gerrit Inspector

Gerrit Inspector is an interactive scriptable environment to inspect and modify internal state of the system.

This environment is available on the system console after the system starts. Leaving the Inspector will shutdown the Gerrit instance.

The environment allows interactive work as well as running of Python scripts for troubleshooting.

Gerrit Inspect can be started by adding '-s' option to the command used to launch the daemon:

  $(bazel info output_base)/external/local_jdk/bin/java \
     -jar bazel-bin/gerrit.war daemon -d ../gerrit_testsite -s
Please refer to this explanation for details why using java -jar isn’t sufficient.

Gerrit Inspector examines Java libraries first, then loads its initialization scripts and then starts a command line prompt on the console:

  Welcome to the Gerrit Inspector
  Enter help() to see the above again, EOF to quit and stop Gerrit
  Jython 2.5.2 (Release_2_5_2:7206, Mar 2 2011, 23:12:06)
  [OpenJDK 64-Bit Server VM (Sun Microsystems Inc.)] on java1.6.0 running for Gerrit 2.3-rc0-163-g01967ef

With the Inspector enabled Gerrit can be used normally and all interfaces (HTTP, SSH etc.) are available.

Care must be taken not to modify internal state of the system when using the Inspector.

Querying the Database

The embedded H2 database can be queried and updated from the command line. If the daemon is not currently running:

  $(bazel info output_base)/external/local_jdk/bin/java \
     -jar bazel-bin/gerrit.war gsql -d ../gerrit_testsite -s
Please refer to this explanation for details why using java -jar isn’t sufficient.

Or, if it is running and the database is in use, connect over SSH using an administrator user account:

  ssh -p 29418 user@localhost gerrit gsql

Debugging JavaScript

When debugging browser specific issues add ?dbg=1 to the URL so the resulting JavaScript more closely matches the Java sources. The debug pages use the GWT pretty format, where function and variable names match the Java sources.


Client-Server RPC

The client-server RPC implementation is gwtjsonrpc, not the stock RPC system that comes with GWT. This buys us automatic XSRF protection. It also makes all of the messages readable and writable by any JSON implementation, facilitating "mashups" and 3rd party clients.

The programming API is virtually identical, except service interfaces extend RemoteJsonService instead of RemoteService.

Why GWT?

We like it. Plus we can write Java code once and run it both in the browser and on the server side.

Google Web Toolkit:

Apache SSHD: