Testing libc++

Getting Started

libc++ uses LIT to configure and run its tests.

The primary way to run the libc++ tests is by using make check-cxx.

However since libc++ can be used in any number of possible configurations it is important to customize the way LIT builds and runs the tests. This guide provides information on how to use LIT directly to test libc++.

Please see the Lit Command Guide for more information about LIT.


After building libc++, you can run parts of the libc++ test suite by simply running llvm-lit on a specified test or directory. If you’re unsure whether the required libraries have been built, you can use the cxx-test-depends target. For example:

$ cd <monorepo-root>
$ make -C <build> cxx-test-depends # If you want to make sure the targets get rebuilt
$ <build>/bin/llvm-lit -sv libcxx/test/std/re # Run all of the std::regex tests
$ <build>/bin/llvm-lit -sv libcxx/test/std/depr/depr.c.headers/stdlib_h.pass.cpp # Run a single test
$ <build>/bin/llvm-lit -sv libcxx/test/std/atomics libcxx/test/std/threads # Test std::thread and std::atomic

If you used ninja as your build system, running ninja -C <build> check-cxx will run all the tests in the libc++ testsuite.


If you used the Bootstrapping build instead of the default runtimes build, the cxx-test-depends target is instead named runtimes-test-depends, and you will need to prefix <build>/runtimes/runtimes-<target>-bins/ to the paths of all tests. For example, to run all the libcxx tests you can do <build>/bin/llvm-lit -sv <build>/runtimes/runtimes-bins/libcxx/test.

In the default configuration, the tests are built against headers that form a fake installation root of libc++. This installation root has to be updated when changes are made to the headers, so you should re-run the cxx-test-depends target before running the tests manually with lit when you make any sort of change, including to the headers. We recommend using the provided libcxx/utils/libcxx-lit script to automate this so you don’t have to think about building test dependencies every time:

$ cd <monorepo-root>
$ libcxx/utils/libcxx-lit <build> -sv libcxx/test/std/re # Build testing dependencies and run all of the std::regex tests

Sometimes you’ll want to change the way LIT is running the tests. Custom options can be specified using the --param <name>=<val> flag. The most common option you’ll want to change is the standard dialect (ie -std=c++XX). By default the test suite will select the newest C++ dialect supported by the compiler and use that. However, you can manually specify the option like so if you want:

$ libcxx/utils/libcxx-lit <build> -sv libcxx/test/std/containers # Run the tests with the newest -std
$ libcxx/utils/libcxx-lit <build> -sv libcxx/test/std/containers --param std=c++03 # Run the tests in C++03

Other parameters are supported by the test suite. Those are defined in libcxx/utils/libcxx/test/params.py. If you want to customize how to run the libc++ test suite beyond what is available in params.py, you most likely want to use a custom site configuration instead.

The libc++ test suite works by loading a site configuration that defines various “base” parameters (via Lit substitutions). These base parameters represent things like the compiler to use for running the tests, which default compiler and linker flags to use, and how to run an executable. This system is meant to be easily extended for custom needs, in particular when porting the libc++ test suite to new platforms.

Using a Custom Site Configuration

By default, the libc++ test suite will use a site configuration that matches the current CMake configuration. It does so by generating a lit.site.cfg file in the build directory from one of the configuration file templates in libcxx/test/configs/, and pointing llvm-lit (which is a wrapper around llvm/utils/lit/lit.py) to that file. So when you’re running <build>/bin/llvm-lit either directly or indirectly, the generated lit.site.cfg file is always loaded instead of libcxx/test/lit.cfg.py. If you want to use a custom site configuration, simply point the CMake build to it using -DLIBCXX_TEST_CONFIG=<path-to-site-config>, and that site configuration will be used instead. That file can use CMake variables inside it to make configuration easier.

$ cmake <options> -DLIBCXX_TEST_CONFIG=<path-to-site-config>
$ libcxx/utils/libcxx-lit <build> -sv libcxx/test # will use your custom config file

Additional tools

The libc++ test suite uses a few optional tools to improve the code quality.

These tools are: - clang-tidy (you might need additional dev packages to compile libc++-specific clang-tidy checks)

Reproducing CI issues locally

Libc++ has extensive CI that tests various configurations of the library. The testing for all these configurations is located in libcxx/utils/ci/run-buildbot. Most of our CI jobs are being run on a Docker image for reproducibility. The definition of this Docker image is located in libcxx/utils/ci/Dockerfile. If you are looking to reproduce the failure of a specific CI job locally, you should first drop into a Docker container that matches our CI images by running libcxx/utils/ci/run-buildbot-container, and then run the specific CI job that you’re interested in (from within the container) using the run-buildbot script above. If you want to control which compiler is used, you can set the CC and the CXX environment variables before calling run-buildbot to select the right compiler. Take note that some CI jobs are testing the library on specific platforms and are not run in our Docker image. In the general case, it is not possible to reproduce these failures locally, unless they aren’t specific to the platform.

Also note that the Docker container shares the same filesystem as your local machine, so modifying files on your local machine will also modify what the Docker container sees. This is useful for editing source files as you’re testing your code in the Docker container.

Writing Tests

When writing tests for the libc++ test suite, you should follow a few guidelines. This will ensure that your tests can run on a wide variety of hardware and under a wide variety of configurations. We have several unusual configurations such as building the tests on one host but running them on a different host, which add a few requirements to the test suite. Here’s some stuff you should know:

  • All tests are run in a temporary directory that is unique to that test and cleaned up after the test is done.

  • When a test needs data files as inputs, these data files can be saved in the repository (when reasonable) and referenced by the test as // FILE_DEPENDENCIES: <path-to-dependencies>. Copies of these files or directories will be made available to the test in the temporary directory where it is run.

  • You should never hardcode a path from the build-host in a test, because that path will not necessarily be available on the host where the tests are run.

  • You should try to reduce the runtime dependencies of each test to the minimum. For example, requiring Python to run a test is bad, since Python is not necessarily available on all devices we may want to run the tests on (even though supporting Python is probably trivial for the build-host).

Structure of a test

Some platforms where libc++ is tested have requirement on the signature of main and require main to explicitly return a value. Therefore the typical main function should look like:

int main(int, char**) {
  return 0;

The C++ Standard has constexpr requirements. The typical way to test that, is to create a helper test function that returns a bool and use the following main function:

constexpr bool test() {
  return true;

int main(int, char**) {

  return 0;

Tests in libc++ mainly use assert and static_assert for testing. There are a few helper macros and function that can be used to make it easier to write common tests.


The header contains several macros with user specified log messages. This is useful when a normal assertion failure lacks the information to easily understand why the test has failed. This usually happens when the test is in a helper function. For example the std::format tests use a helper function for its validation. When the test fails it will give the line in the helper function with the condition out == expected failed. Without knowing what the value of format string, out and expected are it is not easy to understand why the test has failed. By logging these three values the point of failure can be found without resorting to a debugger.

Several of these macros are documented to take an ARG. This ARG:

  • if it is a const char* or std::string its contents are written to the stderr,

  • otherwise it must be a callable that is invoked without any additional arguments and is expected to produce useful output to e.g. stderr.

This makes it possible to write additional information when a test fails, either by supplying a hard-coded string or generate it at runtime.


This macro is an unconditional failure with a log message ARG. The main use-case is to fail when code is reached that should be unreachable.


This macro requires its CONDITION to evaluate to true. If that fails it will fail the test with a log message ARG.


If the library under test is libc++ it behaves like TEST_REQUIRE, else it is a no-op. This makes it possible to test libc++ specific behaviour. For example testing whether the what() of an exception thrown matches libc++’s expectations. (Usually the Standard requires certain exceptions to be thrown, but not the contents of its what() message.)


Validates execution of EXPR does not throw an exception.


Validates the execution of EXPR throws an exception of the type TYPE.


Validates the execution of EXPR throws an exception of the type TYPE which passes validation of PRED. Using this macro makes it easier to write tests using exceptions. The code to write a test manually would be:

void test_excption([[maybe_unused]] int arg) {
#ifndef TEST_HAS_NO_EXCEPTIONS // do nothing when tests are disabled
  try {
    assert(false); // validates foo really throws
  } catch ([[maybe_unused]] const bar& e) {
    LIBCPP_ASSERT(e.what() == what);
  assert(false); // validates bar was thrown

The same test using a macro:

void test_excption([[maybe_unused]] int arg) {
                          [](const bar& e) {
                            LIBCPP_ASSERT(e.what() == what);


This file contains a helper macro TEST_WRITE_CONCATENATED to lazily concatenate its arguments to a std::string and write it to stderr. When the output can’t be concatenated a default message will be written to stderr. This is useful for tests where the arguments use different character types like char and wchar_t, the latter can’t simply be written to stderr.

This macro is in a different header as assert_macros.h since it pulls in additional headers.

Test names

The names of test files have meaning for the libc++-specific configuration of Lit. Based on the pattern that matches the name of a test file, Lit will test the code contained therein in different ways. Refer to the Lit Meaning of libc++ Test Filenames when determining the names for new test files.

Lit Meaning of libc++ Test Filenames

Name Pattern



Checks whether the C++ code in the file compiles, links and runs successfully.


Same as FOO.pass.cpp, but for Objective-C++.


Checks whether the C++ code in the file compiles successfully. In general, prefer compile tests over verify tests, subject to the specific recommendations, below, for when to write verify tests.


Same as FOO.compile.pass.cpp, but for Objective-C++.


Checks that the code in the file does not compile successfully.


Compiles with clang-verify. This type of test is automatically marked as UNSUPPORTED if the compiler does not support clang-verify. For additional information about how to write verify tests, see the Internals Manual. Prefer verify tests over compile tests to test that compilation fails for a particular reason. For example, use a verify test to ensure that

  • an expected static_assert is triggered;

  • the use of deprecated functions generates the proper warning;

  • removed functions are no longer usable; or

  • return values from functions marked [[nodiscard]] are stored.


Checks that the C++ code in the file compiles and links successfully – no run attempted.


Same as FOO.link.pass.cpp, but for Objective-C++.


Checks whether the C++ code in the file fails to link after successful compilation.


Same as FOO.link.fail.cpp, but for Objective-C++.


A builtin Lit Shell test.


A variant of a Lit Shell test that generates one or more Lit tests on the fly. Executing this test must generate one or more files as expected by LLVM split-file. Each generated file will drive an invocation of a separate Lit test. The format of the generated file will determine the type of Lit test to be executed. This can be used to generate multiple Lit tests from a single source file, which is useful for testing repetitive properties in the library. Be careful not to abuse this since this is not a replacement for usual code reuse techniques.

libc++-Specific Lit Features

Custom Directives

Lit has many directives built in (e.g., DEFINE, UNSUPPORTED). In addition to those directives, libc++ adds two additional libc++-specific directives that makes writing tests easier. See libc++-specific Lit Directives for more information about the FILE_DEPENDENCIES, ADDITIONAL_COMPILE_FLAGS, and MODULE_DEPENDENCIES libc++-specific directives.

libc++-specific Lit Directives





// FILE_DEPENDENCIES: file, directory, /path/to/file, ...

The paths given to the FILE_DEPENDENCIES directive can specify directories or specific files upon which a given test depend. For example, a test that requires some test input stored in a data file would use this libc++-specific Lit directive. When a test file contains the FILE_DEPENDENCIES directive, Lit will collect the named files and copy them to the directory represented by the %T substitution before the test executes. The copy is performed from the directory represented by the %S substitution (i.e. the source directory of the test being executed) which makes it possible to use relative paths to specify the location of dependency files. After Lit copies all the dependent files to the directory specified by the %T substitution, that directory should contain all the necessary inputs to run. In other words, it should be possible to copy the contents of the directory specified by the %T substitution to a remote host where the execution of the test will actually occur.


// ADDITIONAL_COMPILE_FLAGS: flag1 flag2 ...

The additional compiler flags specified by a space-separated list to the ADDITIONAL_COMPILE_FLAGS libc++-specific Lit directive will be added to the end of the %{compile_flags} substitution for the test that contains it. This libc++-specific Lit directive makes it possible to add special compilation flags without having to resort to writing a .sh.cpp test (see Lit Meaning of libc++ Test Filenames), more powerful but perhaps overkill.


// MODULE_DEPENDENCIES: std std.compat

This directive will build the required C++23 standard library modules and add the additional compiler flags in %{compile_flags}. (Libc++ offers these modules in C++20 as an extension.)


Libc++ contains benchmark tests separately from the test of the test suite. The benchmarks are written using the Google Benchmark library, a copy of which is stored in the libc++ repository.

For more information about using the Google Benchmark library see the official documentation.

Building Benchmarks

The benchmark tests are not built by default. The benchmarks can be built using the cxx-benchmarks target.

An example build would look like:

$ cd build
$ ninja cxx-benchmarks

This will build all of the benchmarks under <libcxx-src>/benchmarks to be built against the just-built libc++. The compiled tests are output into build/projects/libcxx/benchmarks.

The benchmarks can also be built against the platforms native standard library using the -DLIBCXX_BUILD_BENCHMARKS_NATIVE_STDLIB=ON CMake option. This is useful for comparing the performance of libc++ to other standard libraries. The compiled benchmarks are named <test>.libcxx.out if they test libc++ and <test>.native.out otherwise.

Also See:

Running Benchmarks

The benchmarks must be run manually by the user. Currently there is no way to run them as part of the build.

For example:

$ cd build/projects/libcxx/benchmarks
$ ./algorithms.libcxx.out # Runs all the benchmarks
$ ./algorithms.libcxx.out --benchmark_filter=BM_Sort.* # Only runs the sort benchmarks

For more information about running benchmarks see Google Benchmark.