WebLayer

WebLayer is a high level embedding API to support building a browser.

Unlike src/content, which is only concerned with a sandboxed multi-process web platform, WebLayer includes modern browser features and Google integration. It's the reusable version of Chrome, which might share some portions of the UI and also its support for all the modern HTML5 and browser features (e.g. UI for permissions, autofill, safe browsing etc...).

While it's built on top of src/content, the expectation is that the API will hide the Content API.

Design Goals

1. WebLayer should be powerful enough to build a modern browser. This means:
   • as fast and secure as Chrome
   • includes all the same web platform features as Chrome (including UIs and system integration code)
2. WebLayer embedders should get new web platform features for free (e.g. don't need to keep updating their UIs or code)
3. WebLayer hides subprocesses, so any code that needs to run in the renderer needs to be part of WebLayer. Script injection is generally discouraged for security and performance reasons.

Resources and Documentation

Mailing list: weblayer-dev@chromium.org

Bug tracker: Internals>WebLayer

Directory Structure

public the C++ and Java public API. These are the only files an app should use

shell sample app

test test harnesses and test-only helper code

tools helper scripts

app internal code which runs at the beginning of each process

browser internal code which runs in the browser process

common internal code which runs in the browser and child processes

renderer internal code which runs in the renderer process

utility internal code which runs in the utility process

Setting Up Your Build Environment

If you haven't done this already, you first need to set up an Android build. If you are a Google employee, reach out to weblayer-team@google.com for internal instructions. Otherwise follow the Android build instructions.

Building and Testing

To run the sample app:

    $ autoninja -C out/Default run_weblayer_shell
    $ out/Default/bin/run_weblayer_shell

To run instrumentation tests:

    $ autoninja -C out/Default weblayer_instrumentation_test_apk
    $ out/Default/bin/run_weblayer_instrumentation_test_apk

Note: this may not work on some versions of Android. If you see an error setting the WebView provider when running instrumentation tests, try running the tests using the WebLayer support APK which uses a different loading path:

    $ autoninja -C out/Default weblayer_support_instrumentation_test_apk
    $ out/Default/bin/run_weblayer_support_instrumentation_test_apk

The test script will build and install all necessary APKs.

Running Skew Tests

Make sure you have the following gn arg:

system_webview_package_name = "com.google.android.webview"

The following tests the latest client library against M80:

    $ autoninja -C out/Default weblayer_instrumentation_test_versions_apk
    $ cipd install --root /tmp/M80 chromium/testing/weblayer-x86 m80
    $ out/Default/bin/run_weblayer_instrumentation_test_versions_apk \
       --test-runner-outdir out/Default \
       --client-outdir out/Default \
       --implementation-outdir /tmp/M80/out/Release

Running WPT

To run WPT on android against weblayer do the following:

    $ WPT_TEST=badging/badge-success.https.html # test or directory you want to run, relative to wpt directory.
    $ autoninja -C out/Default weblayer_shell_wpt
    $ out/Default/bin/run_weblayer_shell_wpt \
      -t Default \
      --isolated-script-test-output out/Default/weblayer_out.json \
      --include $WPT_TEST \
      --ignore-browser-specific-expectations \
      --ignore-default-expectations \
      --avd-config tools/android/avd/proto/generic_android28.textpb

The script will write test artifacts to a directory named layout-test-results that is a sibling of --isolated-script-test-output. You can view the results by serving them over HTTP:

    $ python3 -m http.server 8080 -d out/Default/layout-test-results

Then, navigate to <host>:8080/results.html in your browser.

To run against clank:

    $ WPT_TEST=badging/badge-success.https.html # test or directory you want to run, relative to wpt directory.
    $ autoninja -C out/Default chrome_public_wpt
    $ out/Default/bin/run_chrome_public_wpt \
      -t Default \
      --isolated-script-test-output out/Default/clank_out.json \
      --include $WPT_TEST \
      --ignore-browser-specific-expectations \
      --ignore-default-expectations \
      --avd-config tools/android/avd/proto/generic_android28.textpb

The --ignore-browser-specific-expectations --ignore-default-expectations flags will prevent *Expectations files from being loaded, which helps with error messages on test failures. Once a test is fixed, rerun it without those flags to ensure the Expectations files are correct.

To run against linux with wptrunner (same runner we use on android, which runs normal chrome):

    $ WPT_TEST=badging/badge-success.https.html # test or directory you want to run, relative to wpt directory.
    $ autoninja -C out/Default wpt_tests_isolate
    $ out/Default/bin/run_wpt_tests_isolate \
      -t Default \
      --isolated-script-test-output out/Default/chrome_out.json \
      --include $WPT_TEST

To run against linux with run_web_tests (same runner we use on CI, which runs content_shell):

    $ WPT_TEST=badging/badge-success.https.html # test or directory you want to run, relative to wpt directory.
    $ autoninja -C out/Default blink_tests
    $ ./third_party/blink/tools/run_web_tests.py -t Default external/wpt/$WPT_TEST

Passing in -vvvv may be useful if you want to see loads of information about test execution.

A list of known test failures is in WeblayerWPTExpectations. The values between the brackets at the end of each line list the expected result types that test can have. For example, a test marked as “[ Failure ]” is expected to fail, while a test marked as “[ Failure Pass ]” is expected to be flaky.

Any failing tests should be removed from WeblayerWPTExpectations file once fixed.

Tips

While many WPT tests fail due to features not being implemented in WebLayer, some may fail due to Features that aren‘t getting enabled or switches that aren’t getting passed to the test as they would be for Clank. If a test is failing due to a missing Feature, check the test FieldTrial configuration in fieldtrial_testing_config.json. A missing switch could have several causes, but the flags that get passed to the test originate from third_party/wpt_tools/wpt/tools/wpt/run.py.

Telemetry

Telemetry is run against WebLayer, currently on the bot android-pixel2_weblayer-perf.

Telemetry currently only runs on real hardware. Bug 1067712 is for adding support for emulators.

Tricks:

To see the set of stories executed, click on a successful run, search for performance_weblayer_test_suite and click on the json.output link.

Googlers can submit jobs against your own patch using pinpoint. At the time of this writing, logcat is not captured for successful runs (1067024). Submitting a pinpoint run against a patch with a CHECK will generate logcat. For such a run, the logcat is viewable by way of:

1. Click on Id next to task under Test.
2. Expand + (under More Details).
3. Click on link next to Isolated Outputs.
4. Click on test_results.json.
5. Replace gs:// with https://pantheon.corp.google.com/storage/browser.