Basic Precommit


As part of Apache Hadoop's commit process, all patches to the source base go through a precommit test that does some (relatively) light checking to make sure the proposed change does not break unit tests and/or passes some other prerequisites such as code formatting guidelines. This is meant as a preliminary check for committers so that the basic patch is in a known state and for contributors to know if they have followed the project's guidelines. This check, called test-patch, along with a helper program, called smart-apply-patch, may also be used by individual developers to verify a patch prior to sending to the Apache Hadoop QA systems.

Other projects have adopted a similar methodology after seeing great success in the Apache Hadoop model. Some have even gone as far as forking Apache Hadoop's precommit code and modifying it to meet their project's needs.

One of the key facets of Apache Yetus is to bring together all of these forks under a common code base to help software development
as a whole.


test-patch and smart-apply-patch are written in bash for maximum portability. As such, it mostly assumes the locations of commands to be in the file path. However, in many cases, this assumption may be overridden via command line options.

For Solaris and Solaris-like operating systems, the default location for the POSIX binaries is in /usr/xpg4/bin and the default location for the GNU binaries is /usr/gnu/bin.

Base Requirements

test-patch requires these installed components to execute:

Optional Requirements

Features are plug-in based and enabled either individually or collectively on the command line. From there, these are activated based upon tool availability, the languages being tested, etc. The external dependencies of plug-ins may have different licensing requirements than Apache Yetus.

Bug Systems:

Build Tools:

Automation and Isolation:

Unit Test Formats:

Compiler Support:

Language Support, Licensing, and more:

First Steps

A typical local configuration is to have two repositories. One with the code you are working on and another, clean repository. The workflow looks similar to the following:

$ cd <workrepo>
$ git format-patch master > /tmp/patchfile
$ test-patch --basedir=/some/path/testrepo /tmp/patchfile
  1. In your work repo, generate a patch.
  2. Run your patch against the test repository.

--basedir sets the location of the repository to use for testing. The /tmp/patchfile is the patch file that is generated by git format-patch or git diff or any number of other types of input.

Resetting the Repository

After running that command, the test repository will contain the remnants of the build for later debugging. It is inconvenient to instead use:

$ cd <workrepo>
$ git format-patch master > /tmp/patchfile
$ test-patch --basedir=/some/path/testrepo --resetrepo /tmp/patchfile

--resetrepo tells test patch that it can go into destructive mode. Destructive mode will wipe out any changes made to that repository, so use it with care!

Enabling Features

In general, almost all features are enabled via the plug-in framework. To see what has been enabled, use the --list-plugins option:

$ test-patch --list-plugins

You should see output similar to this:

  ant autoconf cmake gradle make maven nobuild
  asflicense author cc checkstyle dupname findbugs hadolint javac javadoc jshint mvnsite pathlen perlcritic pylint rubocop ruby_lint scalac scaladoc shellcheck shelldocs test4tests unitveto whitespace xml yamllint
  briefreport bugzilla github gitlab htmlout jira
  ctest junit tap

This output shows all of the plug-ins that are available as well as what type of plug-in they are.

From this list, the specific plug-ins can be enabled:

$ test-patch --plugins="ant,maven,shellcheck,xml" <other options>

As a short-cut, every plug-in may be enabled via the special 'all' type:

$ test-patch --plugins="all" <other options>

--plugins also allows some basic "arithmetic":

$ test-patch --plugins="all,-checkstyle,-findbugs" <other options>

This will enable all plug-ins for potential usage, except for checkstyle and findbugs.

NOTE: Many examples in this section will use --plugins=all. Users should set the --plugins option as appropriate.

This command will execute basic patch testing against a patch file stored in "filename":

$ cd <your repo>
$ test-patch --plugins=all --dirty-workspace --project=projectname <filename>

The --dirty-workspace flag tells test-patch that the repository is not clean and it is ok to continue. By default, unit tests are not run since they may take a significant amount of time.

To do turn them on, we need to provide the –run-tests option:

$ cd <your repo>
$ test-patch --plugins=all  --dirty-workspace --run-tests <filename>

This is the same command, but now runs the unit tests.

Output Directory

After the tests have run, there is a directory that contains all of the test-patch related artifacts. This is generally referred to as the patch directory. By default, test-patch tries to make something off of /tmp to contain this content. Using the --patch-dir option, one can specify exactly which directory to use. This is helpful for automated precommit testing so that continuous integration systems knows where to look to gather up the output.

For example:

$ test-patch --plugins=all --patch-dir=${WORKSPACE}/patchdir --basedir=${WORKSPACE}/source ${WORKSPACE}/patchfile

… will trigger test-patch to run in fully automated mode, using ${WORKSPACE}/patchdir as its scratch space, ${WORKSPACE}/source as the source repository, and ${WORKSPACE}/patchfile as the name of the patch to test against. This will always run the unit tests, write answers back to bug systems, remove old, stopped/exited Docker containers after 24 hours and images after 1 week, forcibly use --resetrepo, and more.

NOTE: Make sure to add the patch directory to .gitignore if the directory is inside the source tree to avoid deleting it, as test-patch does a git clean to remove untracked files from previous runs.

Build Tool

Out of the box, test-patch will try to figure out which build tool the project uses. But what if you want to override it? The --build-tool option allows a manual setting:

$ test-patch --plugins=all --build-tool=ant (other options)

will tell test-patch to use ant instead of maven to drive the project.

To disable the build tool entirely, use the nobuild setting:

$ test-patch --plugins=all --build-tool=nobuild (other options)

Providing Patch Files

NOTE: More in-depth information may be found in the bugsystems section.


It is a fairly common practice within the Apache community to use Apache's JIRA instance to store potential patches. As a result, test-patch supports providing just a JIRA issue number via the jira plug-in. test-patch will find the last attachment, download it, then process it.

NOTE: test-patch expects the patch files to follow a particular naming convention. For complete details
on the naming convention please refer to patch-naming-conventions

For example:

$ test-patch --plugins=all HADOOP-9905 (other options)

… will process the patch file associated with this JIRA issue.

If the Apache JIRA system is not in use, then override options may be provided on the command line to point to a different JIRA instance.

$ test-patch --plugins=all --jira-issue-re='^PROJECT-[0-9]+$' --jira-base-url='' PROJECT-90

… will process the patch file attached to PROJECT-90 on the JIRA instance located on the server.


test-patch has built-in support for Github via the github plug-in. test-patch supports many forms of providing pull requests to work on:

$ test-patch --plugins=all --github-repo=apache/pig GH:99


$ test-patch --plugins=all


$ test-patch --plugins=all

… will process PR #99 on the apache/pig repo.


test-patch has support for Gitlab via the gitlab plug-in. test-patch supports many forms of providing merge requests to work on:

$ test-patch --plugins=all --gitlab-repo=_a__w_/yetus GL:1


$ test-patch --plugins=all


$ test-patch --plugins=all

… will process MR #3 on the _a__w_/yetus repo.

Generic URLs

Luckily, test-patch supports ways to provide unified diffs via URLs.

For example:

$ test-patch --plugins=all

… will download and process the file.patch from the webserver.

Project-specific Capabilities

Due to the extensible nature of the system, test-patch allows for projects to define project-specific rules which we call personalities. (How to build those rules is covered elsewhere.) There are two ways to specify which personality to use:

Direct Method

$ test-patch --plugins=all --personality=(filename)

This tells test-patch to use the personality in the given file.

Project Method

However, test-patch can detect if it is a personality that is in its "personality" directory based upon the project name:

$ test-patch --plugins=all --project=(project)

Fork Bomb Protection

By default, test-patch will set the user soft limit (ulimit -Su) to a relatively low 1,000 processes (and, on some operating systems with some languages such as Java, threads!). This is to prevent errant processes from eating up all system resources. If this limit is too low, it may be necessary to use the --proclimit option. For example:

$ test-patch --plugins=all --proclimit=10000

… will set it to be 10,000 processes.

NOTE: The actual implementation of this feature is dependent upon the version of Bash. For bash v4 and higher (most operating systems), the fork bomb protection is generally only used for the build and QA tools. This means Apache Yetus should continue to function. For earlier versions of bash (e.g., OS X), the limit is applied to all of test-patch. If the limit is hit, Apache Yetus will itself likely crash.


For many projects, it is useful to test Java code against multiple versions of JDKs at the same time. In combination with the java plug-in, test-patch can do this with the --multijdkdirs option:

$ test-patch --plugins=all --multijdkdirs="/j/d/k/1,/j/d/k/2"

Not all Java tests support this mode, but those that do will now run their tests with all of the given versions of Java consecutively (e.g., javac–the Java compliation test). Tests that do not support MultiJDK mode (e.g., checkstyle, mvn install) will use JAVA_HOME.

NOTE: JAVA_HOME is always appended to the list of JDKs in MultiJDK mode. If JAVA_HOME is in the list, it will be moved to the end.


test-patch also has a built-in mode (i.e., no plug-in required) to utilize Docker:

$ test-patch --docker

This command will do some preliminary setup and then re-execute itself inside a Docker container. For more information on how to provide a custom Dockerfile and other Docker-specific features, see the specific precommit Docker support page and the Apache Yetus Docker Hub Images page for more information on the convenience Docker images.

In Closing

test-patch has many other features and command line options for the basic user. Many of these are self-explanatory. To see the list of options, run test-patch without any options or with --help.