TL;DR

If you have the prerequisites, running

  $ bazel build gerrit

should generate a .war file under bazel-bin/gerrit.war.

Prerequisites

To build Gerrit from source, you need:

  • A Linux or macOS system (Windows is not supported at this time)

  • A JDK for Java 8|11|…​

  • Python 3

  • Node.js (including npm)

  • Bower (npm install -g bower)

  • Bazel -launched with Bazelisk

  • Maven

  • zip, unzip

  • curl

  • gcc

Bazel

Bazelisk is a version manager for Bazel, similar to how nvm manages npm versions. It takes care of downloading and installing Bazel itself, so you don’t have to worry about using the correct version of Bazel. Bazelisk can be installed in different ways: Install

Java

Ensure that the required Java version is installed and that JAVA_HOME is set to it.

To check the installed version of Java, open a terminal window and run:

java -version

Java 8 support (deprecated)

Java 8 is a legacy Java release and support for Java 8 will be discontinued in future gerrit releases. To build Gerrit with Java 8 language level, run:

  $ bazel build --java_toolchain //tools:error_prone_warnings_toolchain :release

Java 11 support

Java language level 11 is the default. To build Gerrit with Java 11 language level, run:

  $ bazel build :release

Java 13 support

Java 13 (and newer) is supported through vanilla java toolchain Bazel option. To build Gerrit with Java 13 and newer, specify vanilla java toolchain and provide the path to JDK home:

  $ bazel build \
    --define=ABSOLUTE_JAVABASE=<path-to-java-13> \
    --javabase=@bazel_tools//tools/jdk:absolute_javabase \
    --host_javabase=@bazel_tools//tools/jdk:absolute_javabase \
    --host_java_toolchain=@bazel_tools//tools/jdk:toolchain_vanilla \
    --java_toolchain=@bazel_tools//tools/jdk:toolchain_vanilla \
    :release

To run the tests, --javabase option must be passed as well, because bazel test runs the test using the target javabase:

  $ bazel test \
    --define=ABSOLUTE_JAVABASE=<path-to-java-13> \
    --javabase=@bazel_tools//tools/jdk:absolute_javabase \
    --host_javabase=@bazel_tools//tools/jdk:absolute_javabase \
    --host_java_toolchain=@bazel_tools//tools/jdk:toolchain_vanilla \
    --java_toolchain=@bazel_tools//tools/jdk:toolchain_vanilla \
    //...

To avoid passing all those options on every Bazel build invocation, they could be added to ~/.bazelrc resource file:

$ cat << EOF > ~/.bazelrc
build --define=ABSOLUTE_JAVABASE=<path-to-java-13>
build --javabase=@bazel_tools//tools/jdk:absolute_javabase
build --host_javabase=@bazel_tools//tools/jdk:absolute_javabase
build --host_java_toolchain=@bazel_tools//tools/jdk:toolchain_vanilla
build --java_toolchain=@bazel_tools//tools/jdk:toolchain_vanilla
EOF

Now, invoking Bazel with just bazel build :release would include all those options.

Node.js and npm packages

Building on the Command Line

Gerrit Development WAR File

To build the Gerrit web application:

  bazel build gerrit

The output executable WAR will be placed in:

  bazel-bin/gerrit.war

Gerrit Release WAR File

To build the Gerrit web application that includes the Gerrit UI, core plugins and documentation:

  bazel build release

The output executable WAR will be placed in:

  bazel-bin/release.war

Headless Mode

To build Gerrit in headless mode, i.e. without the Gerrit UI:

  bazel build headless

The output executable WAR will be placed in:

  bazel-bin/headless.war

Extension and Plugin API JAR Files

To build the extension, plugin and acceptance-framework JAR files:

  bazel build api

The output archive that contains Java binaries, Java sources and Java docs will be placed in:

  bazel-bin/api.zip

Install {extension,plugin,acceptance-framework}-api to the local maven repository:

  tools/maven/api.sh install

Install gerrit.war to the local maven repository:

  tools/maven/api.sh war_install

Plugins

  bazel build plugins:core

The output JAR files for individual plugins will be placed in:

  bazel-bin/plugins/<name>/<name>.jar

The JAR files will also be packaged in:

  bazel-bin/plugins/core.zip

To build a specific plugin:

  bazel build plugins/<name>

The output JAR file will be be placed in:

  bazel-bin/plugins/<name>/<name>.jar

Note that when building an individual plugin, the core.zip package is not regenerated.

Using an IDE.

IntelliJ

The Gerrit build works with Bazel’s IntelliJ plugin. Please follow the instructions on IntelliJ Setup.

Eclipse

Generating the Eclipse Project

Create the Eclipse project:

  tools/eclipse/project.py

and then follow the setup instructions.

Refreshing the Classpath

If an updated classpath is needed, the Eclipse project can be refreshed and missing dependency JARs can be downloaded by running project.py again. For IntelliJ, you need to click the Sync Project with BUILD Files button of Bazel plugin.

Documentation

To build only the documentation for testing or static hosting:

  bazel build Documentation:searchfree

The html files will be bundled into searchfree.zip in this location:

  bazel-bin/Documentation/searchfree.zip

To generate HTML files skipping the zip archiving:

  bazel build Documentation

And open bazel-bin/Documentation/index.html.

To build the Gerrit executable WAR with the documentation included:

  bazel build withdocs

The WAR file will be placed in:

  bazel-bin/withdocs.war

Alternatively, one can generate the documentation as flat files:

  bazel build Documentation:Documentation

The html, css, js files are placed in:

 `bazel-bin/Documentation/`

Running Unit Tests

  bazel test --build_tests_only //...

Debugging tests:

  bazel test --test_output=streamed --test_filter=com.gerrit.TestClass.testMethod testTarget

Debug test example:

  bazel test --test_output=streamed --test_filter=com.google.gerrit.acceptance.api.change.ChangeIT.getAmbiguous //javatests/com/google/gerrit/acceptance/api/change:api_change

To run a specific test group, e.g. the rest-account test group:

  bazel test //javatests/com/google/gerrit/acceptance/rest/account:rest_account

To run SSH tests using JSch ssh client:

  bazel test --test_env=SSH_CLIENT_IMPLEMENTATION=JSCH //...

To run SSH tests using Apache MINA ssh client:

  bazel test --test_env=SSH_CLIENT_IMPLEMENTATION=APACHE //...

To run only tests that do not use SSH:

  bazel test --test_env=GERRIT_USE_SSH=NO //...

To exclude tests that have been marked as flaky:

  bazel test --test_tag_filters=-flaky //...

To exclude tests that require a Docker host:

  bazel test --test_tag_filters=-docker //...

To exclude tests that require very recent git client version:

  bazel test --test_tag_filters=-git-protocol-v2 //...

To ignore cached test results:

  bazel test --cache_test_results=NO //...

To run one or more specific groups of tests:

  bazel test --test_tag_filters=api,git //...

The following values are currently supported for the group name:

  • annotation

  • api

  • docker

  • edit

  • elastic

  • git

  • git-protocol-v2

  • git-upload-archive

  • notedb

  • pgm

  • rest

  • server

  • ssh

Bazel itself supports a multitude of ways to specify targets for fine-grained test selection that can be combined with many of the examples above.

Debugging Unit Tests

In some cases it may be necessary to debug a test while running it in bazel. For example, when we observe a different test result in Eclipse and bazel. Using the --java_debug option will start the JVM in debug mode and await for a remote debugger to attach.

Example:

  bazel test --java_debug --test_tag_filters=delete-project //...
  ...
  Listening for transport dt_socket at address: 5005
  ...

Now attach with a debugger to the port 5005. For example use "Remote Java Application" launch configuration in Eclipe and specify the port 5005.

Elasticsearch

Successfully running the Elasticsearch tests requires Docker, and may require setting the local virtual memory on linux and macOS.

On macOS, if using Docker Desktop, the effective memory value can be set in the Preferences, under the Advanced tab. The default value usually does not suffice and is causing premature container exits. That default is currently 2 GB and should be set to at least 5 (GB).

If Docker is not available, the Elasticsearch tests will be skipped. Note that Bazel currently does not show the skipped tests.

Controlling logging level

Per default, logging level is set to INFO level for all tests. The DEBUG log level can be enabled for the tests.

In IDE, set -Dgerrit.logLevel=debug as a VM argument. With bazel, pass GERRIT_LOG_LEVEL=debug environment variable:

  bazel test --test_filter=com.google.gerrit.server.notedb.ChangeNotesTest \
  --test_env=GERRIT_LOG_LEVEL=debug \
  javatests/com/google/gerrit/server:server_tests

The log results can be found in: bazel-testlogs/javatests/com/google/gerrit/server/server_tests/test.log.

Dependencies

Dependency JARs are normally downloaded as needed, but you can download everything upfront. This is useful to enable subsequent builds to run without network access:

  bazel fetch //...

When downloading from behind a proxy (which is common in some corporate environments), it might be necessary to explicitly specify the proxy that is then used by curl:

  export http_proxy=http://<proxy_user_id>:<proxy_password>@<proxy_server>:<proxy_port>

Redirection to local mirrors of Maven Central and the Gerrit storage bucket is supported by defining specific properties in local.properties, a file that is not tracked by Git:

  echo download.GERRIT = http://nexus.my-company.com/ >>local.properties
  echo download.MAVEN_CENTRAL = http://nexus.my-company.com/ >>local.properties

The local.properties file may be placed in the root of the gerrit repository being built, or in ~/.gerritcodereview/. The file in the root of the gerrit repository has precedence.

Building against unpublished Maven JARs

To build against unpublished Maven JARs, like PrologCafe, the custom JARs must be installed in the local Maven repository (mvn clean install) and maven_jar() must be updated to point to the MAVEN_LOCAL Maven repository for that artifact:

 maven_jar(
   name = 'prolog-runtime',
   artifact = 'com.googlecode.prolog-cafe:prolog-runtime:42',
   repository = MAVEN_LOCAL,
 )

Building against artifacts from custom Maven repositories

To build against custom Maven repositories, two modes of operations are supported: with rewrite in local.properties and without.

Without rewrite the URL of custom Maven repository can be directly passed to the maven_jar() function:

  GERRIT_FORGE = 'http://gerritforge.com/snapshot'

  maven_jar(
    name = 'gitblit',
    artifact = 'com.gitblit:gitblit:1.4.0',
    sha1 = '1b130dbf5578ace37507430a4a523f6594bf34fa',
    repository = GERRIT_FORGE,
 )

When the custom URL has to be rewritten, then the same logic as with Gerrit known Maven repository is used: Repo name must be defined that matches an entry in local.properties file:

  download.GERRIT_FORGE = http://my.company.mirror/gerrit-forge

And corresponding WORKSPACE excerpt:

  GERRIT_FORGE = 'GERRIT_FORGE:'

  maven_jar(
    name = 'gitblit',
    artifact = 'com.gitblit:gitblit:1.4.0',
    sha1 = '1b130dbf5578ace37507430a4a523f6594bf34fa',
    repository = GERRIT_FORGE,
 )

Building against SNAPSHOT Maven JARs

To build against SNAPSHOT Maven JARs, the complete SNAPSHOT version must be used:

 maven_jar(
   name = "pac4j-core",
   artifact = "org.pac4j:pac4j-core:3.5.0-SNAPSHOT-20190112.120241-16",
   sha1 = "da2b1cb68a8f87bfd40813179abd368de9f3a746",
 )

To accelerate builds, several caches are activated per default:

  • ~/.gerritcodereview/bazel-cache/downloaded-artifacts

  • ~/.gerritcodereview/bazel-cache/repository

  • ~/.gerritcodereview/bazel-cache/cas

The downloaded-artifacts cache can be relocated by setting the GERRIT_CACHE_HOME environment variable. The other two can be adjusted with bazel build options --repository_cache and --disk_cache respectively.

Currently none of these caches have a maximum size limit. See this bazel issue for details. Users should watch the cache sizes and clean them manually if necessary.

NPM Binaries

Parts of the Gerrit web app build require running NPM-based JavaScript programs as "binaries". We don’t attempt to resolve and download NPM dependencies at build time, but instead use pre-built bundles of the NPM binary along with all its dependencies. Some packages on registry.npmjs.org come with their dependencies bundled, but this is the exception rather than the rule. More commonly, to add a new binary to this list, you will need to bundle the binary yourself.

Note
We can only use binaries that meet certain licensing requirements, and that do not include any native code.

Start by checking that the license and file types of the bundle are acceptable:

  gerrit_repo=/path/to/gerrit
  package=some-npm-package
  version=1.2.3

  # Note - yarn must be installed before running the following commands
  yarn global add license-checker && \
  rm -rf /tmp/$package-$version && mkdir -p /tmp/$package-$version && \
  cd /tmp/$package-$version && \
  yarn add $package@$version && \
  license-checker | grep licenses: | sort -u

This will output a list of the different licenses used by the package and all its transitive dependencies. We can only legally distribute a bundle via our storage bucket if the licenses allow us to do so. As long as all of the listed license are allowed by Google’s. Any by_exception_only, commercial, prohibited, or unlisted licenses are not allowed; otherwise, it is ok to distribute the source. If in doubt, contact a maintainer who is a Googler.

Next, check the file types:

  cd /tmp/$package-$version
  find . -type f | xargs file | grep -v 'ASCII\|UTF-8\|empty$'

If you see anything that looks like a native library or binary, then we can’t use the bundle.

If everything looks good, install the package with the following command:

# Add to ui_npm. Other packages.json can be updated in the same way
cd $gerrit_repo/polygerrit-ui/app
bazel run @nodejs//:yarn add $package

Update the polygerrit-ui/app/node_modules_licenses/licenses.ts file. You should add licenses for the package itself and for all transitive depndencies. If you forgot to add a license, the Documentation:check_licenses test will fail.

After the update, commit all changes to the repository (including yarn.lock).

Note

If a npm package has transitive dependencies (or just several files) with a not allowed license and you can’t avoid use it in release, then you can add this package. For example some packages contain demo-code with a different license. Another example - optional dependencies, which are not needed to build the Gerrit web app, but they are installed together with the package anyway.

In this case you should exclude all files and/or transitive dependencies with a not allowed license. Adding such package requires additional updates:

  • Add dependencies (or files) to the license.ts with an appropriate license marked with allowed: false.

  • update package.json postinstall script to remove all non-allowed files (if you don’t update postinstall script, Documentation:check_licenses test will fail.)

Update NPM Binaries

To update a NPM binary the same actions as for a new one must be done (check licenses, update licenses.ts file, etc…​). The only difference is a command to install a package: instead of bazel run @nodejs//:yarn add $package you should run the bazel run @nodejs//:yarn upgrade …​ command with correct arguments. You can find the list of arguments in the yarn upgrade doc.

Google Remote Build Support

The Bazel build can be used with Google’s Remote Build Execution.

This needs the following setup steps:

gcloud auth application-default login
gcloud services enable remotebuildexecution.googleapis.com  --project=${PROJECT}

Create a worker pool. The instances should have at least 4 CPUs each for adequate performance.

gcloud alpha remote-build-execution worker-pools create default \
    --project=${PROJECT} \
    --instance=default_instance \
    --worker-count=50 \
    --machine-type=e2-standard-4 \
    --disk-size=200

To use RBE, execute

bazel test --config=remote \
    --remote_instance_name=projects/${PROJECT}/instances/default_instance \
    javatests/...