AOSC Wiki / Developer / Packaging / .

Autobuild3 Testing Framework

Advanced usage of Autobuild3 testing framework and its internal implementation details


Proper package testing is an important and necessary way to assure package quality. Testing, in this case, involves upstream-supplied tests (unit tests and/or integration tests) and distribution tests (which should be implemented by packagers or other AOSC maintainers).

Most of the tests can be done in the packaging building process, in this case, the Autobuild3. The audience of this article is packagers who want to use the package testing framework in Autobuild3. Before we continue, make sure you have read the brief introduction in advanced techniques.


A test in Autobuild3 testing framework refers to a set of operations that supplies a series of test results (passed or not, and why not) to help packagers determine the quality of a package. For a given package, multiple tests can be specified.


A test can be in these states:

A finished test must return its status by return value.

A non-critical test result can give additional information. This may include failed sub-tests, statistics and other machine readable information.

Logs are directly printed to stdout and stderr. They're not part of the test result.

Automatic detection

A package, depending on its ABTYPE, may leverage one of the pre-defined testing templates via automatic detection and generation.

For packages that may leverage a pre-defined testing templates, autobuild/defines can override some of its behavior. To disable the automatically generated test, use ABTEST_AUTO_DETECT=no. To override the anchor point where tests runs in the whole building process, use ABTEST_AUTO_DETECT_ANCHOR. For valid values of ABTEST_AUTO_DETECT_ANCHOR, see anchors.

Execution environments

Execution environment are environments for running the tests. A such environement is an abstraction with:

List of currently supported execution environments:


TESTTYPE defines the testing template to use.

A special one called "custom" can invoke your own test script. Execution of a autobuild/check script is done by this "custom" template.

Multi-case testing

For more complex packages with multiple tests or other requirements, you may set up multi-case testing as follows.

These tests should be specified by package metadata file called test specification file under tests/ directory. The metadata file name is the test name. Different tests in a same package must not share a test name.

Test data consist of self-written scripts, example file for the software to process, or standard outputs or screenshots for comparison. These data should be placed in autobuild/testdata. Large files, copyrighted assets, and other controversial files should be hosted elsewhere, keeping the aosc-os-abbs repository clean. Placing files by test name is recommended.

To enable these tests, put ABTESTS="foo bar" in the defines file. ABTESTS is a space separated string, packagers should enable needed tests by list them in this variable.



Test Metadata

Except for the automatically generated test, tests should have their properties placed in the test specification file.


A human readable short description of the test. Optional but strongly recommended.


A space separated list of dependencies for running the test. Its function and syntax is the same as PKGDEP, except that these packages only get installed if the test it belongs to is enabled. Defaults to "".


The execution environment name. Defaults to "plain".


The test type. Required.

Test type dependent metadata

Internal Implementation

The Autobuild3 testing framework consists of these parts:


Autobuild3 reads the ABTESTS variable in defines. For each test name in ABTESTS, create a bash sub-shell that sources the test specification file. The sub-shell will output needed variables to its stdout. Then the main process can read the output, so that each test's main properties will be known by the main process. These variables should be named as ABTEST_${TEST_NAME}_${VARIABLE_NAME}.

Involves proc/ and lib/


Tests may run at different stage of the build process. Most likely we'd run tests after compilation. To specify when to run the test, we inserted some anchor scripts in proc/ directory. In custom tests, we will use CUSTOM_ANCHOR for specifying this. Anchor scripts are placed in the proc/ directory like other build procedures, they invokes corresponding execution environment and test scripts accordingly.

Involves proc/, proc/ and lib/

Execution wrappers

Autobuild3 runs test as a new process, in contrib/autobuild-test. This design provides the ability to run tests in other environments and even hosts or run tests without rebuilding the whole project. You can also manually run autobuild test <testname> if you know what you are doing.

Autobuild3's testing implementation leverages only parts of Autobuild3's functionalities. Only basic proc/ scripts and essential libs are loaded. Not all of the variable are available. Users should always be aware that they are isolated by processes or even virtual machines. The communication are limited to the defined test result reporting mechanism and build directory reading.

Test scripts

Test scripts defines the actual test procedures. They looks and works just like a normal build script under ab3/build/.

As described in the execution wrappers, the results are passed via temporary files and return values. If you're writing custom test scripts, keep an eye on abtest_result function for additional and always return a meaningful exit code.

Involves lib/, tests/ and contrib/autobuild-test.


Collectors copies test result to dpkg control file and generates a machine-readable result for future use such as quality assurance.

Involves pm/dpkg/pack.