Emscripten has a comprehensive test suite, which covers virtually all Emscripten functionality. These tests are an excellent resource for developers as they provide practical examples of most features, and are known to pass on the master branch. In addition to correctness tests, there are also benchmarks that you can run.
This article explains how to run the test and benchmark suite, and provides an overview of what tests are available.
To run the tests, you need an emscripten setup, as it will run
emcc and other
commands. See the developer’s guide for
how best to do that.
Run the test suite runner (tests/runner) with no arguments to see the help message:
The tests are divided into modes. You can run either an entire mode or an individual test, or use wildcards to run some tests in some modes. For example:
# run one test (in the default mode) tests/runner test_loop # run one test in a specific mode (here, asm.js -O2) tests/runner asm2.test_loop # run a test in a bunch of modes (here, all asm.js modes) tests/runner asm*.test_loop # run a bunch of tests in one mode (here, all i64 tests in wasm -O3) tests/runner wasm3.test_*i64* # run all tests in a specific mode (here, asm.js -O1) tests/runner asm1
The core test modes (defined at the bottom of tests/test_core.py) let you run a specific test in either asm.js or wasm, and with different optimization flags. There are also non-core test modes, that run tests in more special manner (in particular, in those tests it is not possible to say “run the test with a different optimization flag” - that is what the core tests are for). The non-core test modes include
other: Non-core tests running in the shell.
browser: Tests that run in a browser.
sockets: Networking tests that run in a browser.
interactive: Browser tests that are not fully automated, and require user interaction (these should be automated eventually).
sanity: Tests for emscripten setting itself up. This modifies your .emscripten file temporarily.
benchmark: Runs benchmarks, measuring speed and code size.
The wildcards we mentioned above work for non-core test modes too, for example:
# run one browser test tests/runner browser.test_sdl_image # run all SDL2 browser tests tests/runner browser.test_sdl2* # run all browser tests tests/runner browser
An individual test can be skipped by passing the “skip:” prefix. E.g.
tests/runner other skip:other.test_cmake
Wildcards can also be passed in skip, so
tests/runner browser skip:browser.test_pthread_*
will run the whole browser suite except for all the pthread tests in it.
You can run a random subset of the test suite, using something like
100 with another number as you prefer. This will run that number of random tests, and tell you the statistical likelihood of almost all the test suite passing assuming those tests do. This works just like election surveys do - given a small sample, we can predict fairly well that so-and-so percent of the public will vote for candidate A. In our case, the “candidates” are pass or fail, and we can predict how much of the test suite will pass given that sample. Assuming the sample tests all pass, we can say with high likelihood that most of the test suite will in fact pass. (Of course, this is no guarantee, and even a single test failure is serious, however, this gives a quick estimate that your patch does not cause significant and obvious breakage.)
Please see the bottom the file tests/test_core.py for the current test modes, as they may change slowly over time. When you want to run the entire test suite locally, these are currently the important commands:
# Run all core wasm tests tests/runner wasm* # Run "other" test suite tests/runner other # Run "browser" test suite - this requires a web browser tests/runner browser # Run "sockets" test suite - this requires a web browser too tests/runner sockets # Run "sanity" test suite - this tests setting up emscripten during # first run, etc., and so it modifies your .emscripten file temporarily. tests/runner sanity # Optionally, also run benchmarks to check for regressions tests/runner benchmark
Emscripten has a benchmark suite that measures both speed and code size, which includes several interesting real-world codebases, from physics engines to compression libraries to virtual machines. It also includes some existing benchmarks such as CoreMark and LINPACK. See for example this post’s section on speed which gives an overview.
To run the benchmark suite, do:
# Run all benchmarks tests/runner benchmark
As with all the test suites, you can also run a specific benchmark:
# Run one specific benchmark tests/runner benchmark.test_skinning
Usually you will want to customize the in tests/test_benchmark.py to run the benchmarks you want (there is currently no external config file). Things you may want to modify include:
benchmarkers is the list of VMs to run the benchmarks on.
DEFAULT_ARG is how long the benchmark should run (they all try to run for
a similar amount of time for consistency).
TEST_REPS is how many times to repeat each run (more will take longer, but
should have less noise).
PROFILING controls whether the builds are set up for profiling (which can
increase code size, so it’s not done by default).
Setting the Debug mode (EMCC_DEBUG) is useful for debugging tests, as it emits debug output and intermediate files (the files go in /tmp/emscripten_temp/):
# On Windows, use "set" to set and un-set the EMCC_DEBUG environment variable: set EMCC_DEBUG=1 tests/runner test_hello_world set EMCC_DEBUG=0 # On Linux, you can do this all in one line EMCC_DEBUG=1 tests/runner test_hello_world # EMCC_DEBUG=2 generates additional debug information. EMCC_DEBUG=2 tests/runner test_hello_world
You can also specify
--save-dir to save the temporary directory that the
test runner uses into /tmp/emscripten_test/. This is a test suite-specific
feature, and is useful for inspecting test outputs as well as temporary files
generated by the test. By default, the temporary directory will be cleaned
between each test run, but you can add
--no-clean to avoid this.
The Debugging topic provides more guidance on how to debug Emscripten-generated code.