There are two suites of tests in the Android Runtime (ART):
ART run-tests are tests exercising the runtime using Dex bytecode. They are written in Java and/or Smali (compiled/assembled as Dex bytecode) and sometimes native code (written as C/C++ testing libraries). Some tests also make use of the Jasmin assembler or the ASM bytecode manipulation tool. Run-tests are executed on the ART runtime (dalvikvm
), possibly preceded by a pre-optimization of the Dex code (using dex2oat
).
The run-tests are identified by directories in this test
directory, named with a numeric prefix and containing an info.txt
file. For most run tests, the sources are in the src
subdirectory. Sources found in the src2
directory are compiled separately but to the same output directory; this can be used to exercise "API mismatch" situations by replacing class files created in the first pass. The src-ex
directory is built separately, and is intended for exercising class loaders. Resources can be stored in the res
directory, which is distributed together with the executable files.
The run-tests logic lives in the test/run-test
Bash script. The execution of a run-test has three main parts: building the test, running the test, and checking the test's output. By default, these three steps are implemented by three Bash scripts located in the test/etc
directory (default-build
, default-run
, and default-check
). These scripts rely on environment variables set by test/run-test
.
The default logic for all of these these steps (build, run, check) is overridden if the test's directory contains a Bash script named after the step (i.e. build
, run
, or check
). Note that the default logic of the "run" step is actually implemented in the "JAR runner" (test/etc/run-test-jar
), invoked by test/etc/default-run
.
After the execution of a run-test, the check step's default behavior (implemented in test/etc/default-check
) is to compare its standard output with the contents of the expected-stdout.txt
file contained in the test's directory; any mismatch triggers a test failure.
The test/run-test
script handles the execution of a single run-test in a given configuration. The Python script test/testrunner/testrunner.py
is a convenient script handling the construction and execution of multiple tests in one configuration or more.
To see the invocation options supported by run-test
and testrunner.py
, run these commands from the Android source top-level directory:
art/test/run-test --help
art/test/testrunner/testrunner.py --help
ART gtests are written in C++ using the Google Test framework. These tests exercise various aspects of the runtime (the logic in libart
, libart-compiler
, etc.) and its binaries (dalvikvm
, dex2oat
, oatdump
, etc.). Some of them are used as unit tests to verify a particular construct in ART. These tests may depend on some test Dex files and core images.
ART gtests are defined in various directories within the ART project (usually in the same directory as the code they exercise). Their source files usually end with the suffix _test.cc
. The construction logic of these tests is implemented in ART's build system (Android.bp
and Android*.mk
files). On host, these gtests can be run by executing m test-art-host-gtest
. On device, the recommended approach is to run these tests in a chroot environment (see README.chroot.md
in this directory).
All tests in either suite can be run using the art/test.py
script. Additionally, run-tests can be run individually. All of the tests can be run on the build host, on a USB-attached device, or using the build host "reference implementation".
ART also supports running target (device) tests in a chroot environment (see README.chroot.md
in this directory). This is currently the recommended way to run tests on target (rather than using art/test.py --target
).
To see command flags run:
$ art/test.py -h
In general all tests require some dependencies to be built before they can be run. In general you can pass the --build-dependencies
flag (also available as short option -b) to art/test.py
program to automatically build required dependencies. One can also directly use the various test-art-...-dependencies
targets listed below.
$ # Build test files $ m test-art-host-run-test-dependencies $ # Run the tests $ art/test.py --host
Or:
$ art/test.py -b --host
$ # Build test files $ m test-art-target-run-test-dependencies $ # Run the tests $ art/test.py --target
Or:
$ art/test.py -b --target
$ art/test.py --host -g
$ art/test.py --target -g
$ # Build test files $ m test-art-host-run-test-dependencies $ art/test.py --host -r
Or:
$ art/test.py -b --host -r
$ art/test.py --target -r
$ # Build test files $ m test-art-host-run-test-dependencies $ # Run the tests $ art/test.py --host -r -t 001-HelloWorld
Or:
$ art/test.py -b --host -r -t 001-HelloWorld
$ art/test.py --target -r -t 001-HelloWorld
$ find out/host/ -type f -name art_runtime_tests # Find the path of the test. $ out/host/linux-x86/nativetest/art_runtime_tests/art_runtime_tests
Add "--no_isolate" to run the tests one by one in single process (disable forking). Add "--gtest_filter=..." to select specific sub-test(s) to run. Prefix by "gdb --args " to run the test in gdb.
Both ART run-tests and gtests are run continuously as part of ART's continuous integration. In addition, two other test suites are run continuously on this service: Libcore tests and JDWP tests.