wiki:NewRunWebKitTests
Last modified 21 months ago Last modified on 04/06/13 16:38:15

new-run-webkit-tests

new-run-webkit-tests (also known as NRWT) is a replacement for the original run-webkit-tests script that was written in Perl. The new script is written in Python and leverages much of the extensive Python infrastructure we have built up over the intervening years. It has a much more extensive unit test infrastructure and will hopefully be easier to maintain than the old script. (Time will tell.)

In addition, NRWT has the ability to run the layout tests in parallel using multiple DumpRenderTrees (or WebkitTestRunners), much stronger features for reporting test failures and managing expected failures, and integration with various dashboards.

This page describes how to run NRWT.

See also

Introduction

NRWT is designed to be run both directly by developers working in a terminal window and by the buildbots. It supports a large number of command line options to customize the behavior and the output; only a few of the most commonly used options are mentioned here.

The complete list of options is given by new-run-webkit-tests --help and new-run-webkit-tests --help-printing.

Basic Usage

NRWT uses the concept of a "platform" to determine which WebKit port / implementation you are attempting to test. Where possible, it picks a default based on the operating system you are running on and the default configuration you have set (if any). However, sometimes you have to give it hints.

By default, when run interactively in a terminal or console window, NRWT attempts to be fairly terse. It will display one line of output on screen, and update that in place with events as the test run occurs. If unexpected things happen (e.g., tests fails that aren't expected to fail), they are printed out as they happen, and then summarized again at the end. If everything passes, it will print a message saying so. Also, it returns the number of unexpected test failures in the exit code for the process.

Examples

Starting up initially

% new-run-webkit-tests
Collecting tests ...

(this may take a few seconds, during which time other message may be displayed, like "checking build", "parsing expectations", etc. will be displayed.

Eventually the tests will start running, and you'll get something resembling a progress meter:

% new-run-webkit-tests
Testing (3%): 863 ran as expected, 0 didn't, 24080 left

and when the test completes, you'll get output something like:

% new-run-webkit-tests
  animations/animation-direction-normal.html -> unexpected text diff mismatch
  compositing/reflections/nested-reflection-animated.html -> unexpected text diff mismatch
                                                         
Retrying 2 unexpected failure(s) ...

  animations/animation-direction-normal.html -> unexpected text diff mismatch
  compositing/reflections/nested-reflection-animated.html -> unexpected text diff mismatch
                                                         
24894 tests ran as expected, 2 didn't:

Regressions: Unexpected text diff mismatch : (10)
  animations/animation-direction-normal.html = TEXT
  compositing/reflections/nested-reflection-animated.html = TEXT

% echo $?
2
%

In addition, by default, a web browser will be opened to an HTML page displaying the tests that failed, and links to the output from the tests. You can notice a few things about this test output:

  1. Tests that fail are automatically retried, to see if the failure was "flaky" and only happens some of the time. A flaky failure (one that fails then passing during the retry is reported on screen, but does not count as an actual failure.
  2. The retries are all done serially in a single thread, to avoid any load-related or ordering-related issues that might be contributing to the flakiness.
  3. The test that failed and the type of test failure that happened are displayed in such a way as to be directly copied into the TestExpectations file.

Commonly used Command Line Flags

NRWT supports a lot of flags. Here are some of the most popular:

--platform the platform to run (e.g., "mac", "win-wk2", etc.
--debug use the Debug configuration
--release use the Release configuration
-n --dry-run dry-run ... do everything but actually run the tests
-p --pixel-tests run with pixel-tests enabled
--no-new-test-results don't create any new test results for new tests (NRWT will create them by default
--no-show-results don't display the summary HTML file in a web browser at the end of the run

A recommended set of command-line flags

dpranke@chromium.org has spent a lot of time hacking on NRWT. The command line argument he most frequently uses are:

% new-run-webkit-tests --clobber-results --no-new-test-results --no-show-results --no-retry-failures

Generating rebaselines into a specific platform path

It may be desired to generate results using a given platform and expect it to be the same in some variations of that same platform, like in Qt which have qt-5.0-wk1, qt-5.0-wk2, and so many other options. The following line would run tests using Qt5 and WK2 but instead of placing the results onto that specific folder, it'll place the results into qt-5.0-wk1 folder:

% run-webkit-tests --qt -2 LayoutTests/canvas --new-baseline --add-platform-exceptions --additional-platform-directory=/opt/git/webkit/LayoutTests/platform/qt-5.0-wk1

The trick is to use --new-baseline so you will replace any existing results for the new ones you generate, and then specifying the platform directory with flags --add-platform-exceptions and --additional-platform-directory.