tree: 46ad91ac02ef79d501e9edf39e1d1ca0aec634ff [path history] [tgz]
  1. src/
  2. BUILD.gn
  3. OWNERS
  4. README.md
tools/test_list_tool/README.md

test_list_tool: the test-list generator

test_list_tool is a host tool that takes tests.json as input and generates a new file called test-list.json. This file contains metadata about the tests, and can include data scraped from generated component manifest files.

Schema

See the test-list crate for the latest test-list.json schema.

Tags

This tool extends the test-list.json schema with a specific set of tags that will be present for all tests in the list. The specific tags are as follows:

KeyDescriptionSource
osThe operating system the test is intended to run on. For example: “linux” or “fuchsia”Copied from tests.json, which is output by the build.
cpuThe CPU this test is intended to run on. For example: “x64” or “arm64”.Copied from tests.json, which is output by the build.
scopeThe scope of the test. For example: “unit” or “integration”.See description below this table.
realmThe test realm that the on-device Test Manager will run this test in. For example: “hermetic” or “system”.Copied from the “fuchsia.test” facet of the test's component manifest file.
hermetic“true” if this test is hermetic, “false” if not hermetic, empty if not applicable.A test is hermetic if its realm is “hermetic”.
legacy_test“true” if this test is a legacy test, empty otherwise.A test is a legacy test if its component URL ends in “.cmx”

Test Scopes

This section describes how we categorize a Test's Scope, which tells us if a test is treated as a “unit test,” “integration test,” “system test,” etc.

The valid scopes supported by this tool are:

ScopeHeuristic
hostThe test points to a host binary, not a Fuchsia package.
host_shellThe test points to a host shell binary, not a Fuchsia package.
unitThe test runs in a hermetic realm and uses a generated default manifest.
integrationThe test runs in a hermetic realm and uses a non-generated custom manifest.
systemThe test runs in a non-hermetic realm.
wrapped_legacyThe test uses the legacy_test_runner and wraps a v1 test. These tests are not hermetic.
fuzzerThe test uses the “fuzzer_package” build rule, and it is hermetic.
prebuiltThe test uses the “prebuilt_test_package” build rule and does not run in a non-hermetic realm.
bootfsThe test is an ELF binary executed out of /boot on the device.
unknownThe test uses an unknown build rule (see algorithm below).
uncategorizedThe test does not fall into any of the above categories.

We apply these heuristics as follows:

  1. Check if the test's name begins with host_ or linux_. These are host tests.
    • Justification: Host test names in fuchsia.git always begin with host_ and no other tests do. There are some tests for Linux that start with linux_, and those tests also are scoped to the host.
  2. Check if the test's name ends with _host_test.sh. These are host_shell tests.
    • Justification: Some host test names end with this suffix, and they are executed as shell scripts on the host.
  3. If the build rule that generated the test is not listed in the following section , the test type is unknown.
    • Justification: We do not want to categorize tests using other rules just yet. This leaves us an opportunity to separately handle other types of tests without miscategorizing them up front.
  4. If the test does not run in the hermetic realm it is a system test.
    • Justification: This realm enforces hermetic execution of tests, and all other tests can change and depend on global system state.
  5. If the test used the fuzzer_package build rule, it is a fuzzer test.
    • Justification: Fuzzers typically have a mode of execution where they complete a single iteration of fuzzing with fixed input to ensure the fuzzer setup itself is bug free. These types of tests are treated separately from general unit and integration tests to track the impact they have on our build.
  6. If the test used the prebuilt_test_package build rule, it is a prebuilt test.
    • Justification: Prebuilt tests are rolled into this repository from an external source and run on the Fuchsia platform defined by this repository. We want to track how many tests are doing this, and their overall impact on test health.
  7. If the test used the bootfs_test build rule, it is a bootfs test.
    • Justification: Bootfs tests are simply ELF binaries run as processes directly out of /boot on the device. We want to track how many tests do this.
  8. If the test uses a generated manifest, it is a unit test.
    • Justification: The generated manifest does not provide the ability to start additional components or processes, so the entire test is contained in a single process.
  9. If the test uses an explicit manifest, it is an integration test.
    • Justification: The primary reason to use a non-generated manifest is to start other components as in the scope of a test. Tests that start and communicate between multiple components are integration tests by definition.

The supported build rules are:

  • fuchsia_unittest_package - Convenience rule for wrapping a single test in a package with a generated manifest.
  • fuchsia_test_package - Rule for including one or more test components in a package.
  • fuchsia_test - Rule for wrapping a single test component in a package.
  • fuzzer_package - Rule for defining a fuzzer and associated test.
  • prebuilt_test_package - Rule for wrapping a prebuilt test binary from another repository as a test package.