Skip to content

Latest commit

 

History

History
171 lines (122 loc) · 7.83 KB

Testing.md

File metadata and controls

171 lines (122 loc) · 7.83 KB
Raw

Testing Ladybird

Tests are locates in Tests/, with a directory for each library.

Every feature or bug fix added to LibWeb should have a corresponding test in Tests/LibWeb. The test should be either a Text, Layout, Ref, or Screenshot test depending on the feature. Tests of internal C++ code go in their own TestFoo.cpp file in Tests/LibWeb.

Running Tests

Note

To reproduce a CI failure, see the section on Running with Sanitizers.

The easiest way to run tests is to use the ladybird.sh script. The LibWeb tests are registered with CMake as a test in UI/CMakeLists.txt. Using the built-in test filtering, you can run all tests with Meta/ladybird.sh test or run just the LibWeb tests with Meta/ladybird.sh test LibWeb. The second way is to invoke the headless browser test runner directly. See the invocation in UI/CMakeLists.txt for the expected command line arguments.

A third way is to invoke ctest directly. The simplest method is to use the default preset from CMakePresets.json:

cmake --preset default
cmake --build --preset default
ctest --preset default

If you want to avoid building and running LibWeb tests, you can use a Lagom-only build.

cmake -GNinja -S Meta/Lagom -B Build/lagom

The tests can be run via ninja after doing a build. Note that test-js requires the LADYBIRD_SOURCE_DIR environment variable to be set to the root of the ladybird source tree.

# /path/to/ladybird repository
export LADYBIRD_SOURCE_DIR=${PWD}
cd Build/lagom
ninja
ninja test

To see the stdout/stderr output of failing tests, the recommended way is to set the environment variable CTEST_OUTPUT_ON_FAILURE to 1.

CTEST_OUTPUT_ON_FAILURE=1 ninja test

# or, using ctest directly...
ctest --output-on-failure

Running with Sanitizers

CI runs host tests with Address Sanitizer and Undefined Sanitizer instrumentation enabled. These tools catch many classes of common C++ errors, including memory leaks, out of bounds access to stack and heap allocations, and signed integer overflow. For more info on the sanitizers, check out the Address Sanitizer wiki page, or the Undefined Sanitizer documentation from clang.

Note that a sanitizer build will take significantly longer than a non-sanitizer build, and will mess with caches in tools such as ccache. The sanitizers can be enabled with the -DENABLE_FOO_SANITIZER set of flags.

The simplest way to enable sanitizers is to use the Sanitizer preset.

cmake --preset Sanitizer
cmake --build --preset Sanitizer
ctest --preset Sanitizer

Or from a Lagom build:

To ensure that the test behaves the same way as CI, make sure to set the ASAN_OPTIONS and UBSAN_OPTIONS appropriately. The Sanitizer test preset already sets these environment variables.

export ASAN_OPTIONS='strict_string_checks=1:check_initialization_order=1:strict_init_order=1:detect_stack_use_after_return=1:allocator_may_return_null=1'
export UBSAN_OPTIONS='print_stacktrace=1:print_summary=1:halt_on_error=1'
cmake -GNinja -S Meta/Lagom -B Build/lagom -DENABLE_ADDRESS_SANITIZER=ON -DENABLE_UNDEFINED_SANITIZER=ON
cd Build/lagom
ninja
CTEST_OUTPUT_ON_FAILURE=1 LADYBIRD_SOURCE_DIR=${PWD}/../.. ninja test

Running the Web Platform Tests

The Web Platform Tests can be run with the WPT.sh script. This script can also be used to compare the results of two test runs.

Enabling the Qt chrome is recommended when running the Web Platform Tests on MacOS. This can be done by running the following command:

cmake -GNinja Build/release -DENABLE_QT=ON

Example usage:

# Run the WPT tests then run them again, comparing the results from the two runs
./Meta/WPT.sh run --log expectations.log css
git checkout my-css-change
./Meta/WPT.sh compare --log results.log expectations.log css
# Pull the latest changes from the upstream WPT repository
./Meta/WPT.sh update
# Run all of the Web Platform Tests, outputting the results to results.log
./Meta/WPT.sh run --log results.log

Importing Web Platform Tests

You can import certain Web Platform Tests (WPT) tests into your Ladybird clone (if they’re tests of type that can be imported — and especially if any code changes you’re making cause Ladybird to pass any WPT tests it hasn’t yet been passing). Here’s how:

./Meta/WPT.sh import html/dom/aria-attribute-reflection.html

That is, you give ./Meta/WPT.sh import the path part of any http://wpt.live/ URL for a WPT test you want to import. It will then download both that test and any of its JavaScript scripts, copy those to the Tests/LibWeb/<test-type>/input/wpt-import directory, run the test, and then in the Tests/LibWeb/<test-type>/expected/wpt-import directory, it will create a file with the expected results from the test.

Writing tests

Running Tests/LibWeb/add_libweb_test.py your-new-test-name will create a new test HTML file in Tests/LibWeb/Text/input/your-new-test-name.html with the correct boilerplate code for a Text test — along with a corresponding expectations file in Tests/LibWeb/Text/expected/your-new-test-name.txt.

After you update/replace the generated boilerplate in your your-new-test-name.html test file with your actual test, running ./Meta/ladybird.sh run headless-browser --run-tests "${LADYBIRD_SOURCE_DIR}/Tests/LibWeb" --rebaseline -f Text/input/foobar.html will regenerate the corresponding expectations file — to match the actual output from your updated test (where /opt/ladybird should be replaced with the absolute path your ladybird clone in your local environment).

Future versions of the add_libweb_test.py script will support other test types.

Text tests

Text tests are intended to test Web APIs that don't have a visual representation. They are written in JavaScript and run in a headless browser. Each test has a test function in a script tag that exercises the API and prints expected results using the println function. println calls are accumulated into an output test file, which is then compared to the expected output file by the test runner.

Text tests can be either sync or async. Async tests should use the done callback to signal completion. Async tests are not necessarily run in an async context, they simply require the test function to signal completion when it is done. If an async context is needed to test the API, the lambda passed to test can be async.

Layout

Layout tests compare the layout tree of a page with an expected one. They are best suited for testing layout code, but are also used for testing some other features that have an observable effect on the layout. No JavaScript is needed — once the page loads, the layout tree will be dumped automatically.

Ref

Reference or "ref" tests compare a screenshot of the test page with one of a reference page. The test passes if the two are identical. These are ideal for testing visual effects such as background images or shadows. If you're finding it difficult to recreate the effect in the reference page, (such as for SVG or canvas,) consider using a Screenshot test instead.

Each Ref test includes a special <link rel="match" href="../expected/my-test-ref.html" /> tag, which the test runner uses to locate the reference page. In this way, multiple tests can use the same reference.

Screenshot

Screenshot tests can be thought of as a subtype of Ref tests, where the reference page is a single <img> tag linking to a screenshot of the expected output. In general, try to avoid using them if a regular Ref test would do, as they are sensitive to small rendering changes, and won't work on all platforms.

Like Ref tests, they require a <link rel="match" href="../expected/my-test-ref.html" /> tag to indicate the reference page to use.