The test harness which runs on our bots (called “runtests”) picks up all executables in the “/boot/test” and “/system/test” directories and runs them. If you provide a command-line argument, such as runtests -S -m widget_test, runtests will only run the single test requested -- in this case, widget_test.

“runtests” takes command-line arguments to toggle classes of tests to execute.

These classes are the following:

  • Small: Isolated tests for functions and classes. These must be totally synchronous and single-threaded. These tests should be parallelizable; there shouldn't be any shared resources between them.
  • Medium: Single-process integration tests. Ideally these are also synchronous and single-threaded but they might run through a large chunk of code in each test case, or they might use disk, making them a bit slower.
  • Large: Slow, multi-process, or particularly incomprehensible single-process integration tests. These tests are often too slow / flaky to run in a CQ, and we should try to limit how many we have.
  • Performance: Tests which are expected to pass, but which are measured using other metrics (thresholds, statistical techniques) to identify regressions.

Since runtests doesn't really know what “class” is executing when it launches a test, it encodes this information in the environment variable RUNTESTS_TEST_CLASS, which is detailed in the unittest header , and lets the executable itself decide what to run / not run. This environment variable is a bitmask indicating which tests to run.

For example, if a a test executable is run with “small” and “medium” tests, it will be executed ONCE with RUNTESTS_TEST_CLASS set to 00000003 (the hex bitwise OR of “TEST_SMALL” and “TEST_MEDIUM” -- though this information should be parsed using the unittest header, as it may be updated in the future).

Magenta Tests (ulib/test, and/or using ulib/unittest)

The following macros can be used to filter tests into these categories:


The legacy RUN_TEST(widget_test) is aliased to mean the same thing as RUN_TEST_SMALL.

Fuchsia Tests (not using ulib/unittest)

The environment variable RUNTESTS_TEST_CLASS will still be available to all executables launched by runtests. The unittest header can be used to parse different categories of tests which the runtests harness attempted to run.

Runtests CLI

By default, runtests will run both small and medium tests.

To determine how to run a custom set of test categories, run runtests -h, which includes usage information.