Managing Test Expectations with new-run-webkit-tests and old-run-webkit-tests

Contents

• How we manage tests that fail
• Suppressing failures using NRWT: the test_expectations.txt file
• Suppressing failures using ORWT: Skipped files and checked-in files.

See also

• NewRunWebKitTests
• Triaging test failures

How we manage tests that fail

The primary function of the LayoutTests is as a *regression test suite*. This means that, while we care about whether a page is being rendered correctly, we care more about whether the page is being rendered the way we expect it to. In other words, we look more for changes in behavior than we do for correctness.

All layout tests have "expected results", which may be one of several forms. The test may produce a text file containing javascript log messages, or a text rendering of the Render Tree. It may also produce a screen capture of the rendered page (if you are running with --pixel-tests enabled) as PNG files as well. For WebAudio tests, we can produce WAV files instead of either text or PNG files. For any of these types of tests, there are files checked into the LayoutTests directory named "-expected.{txt,png,wav}". In many (most?) cases, the output is expected to be generic and match on any webkit port.

When the output doesn't match, there are two potential reasons for it:

1. The port is performing "correctly", but the output simply won't match the generic version. The usual reason for this is for things like form controls, which are rendered differently on each platform.
2. The port is performing "incorrectly" (i.e., the test is failing).

In the former case, the convention is to check in a platform-specific "-expected" file that overrides the generic one.

In the latter case, this is dealt with differently on different ports.

In all ports except for the Chromium ones, the convention is to check in the incorrect output as a platform-specific file, and then file a bug to track the incorrectness. For some tests, on some ports, the test is *never* expected to pass, in which case the test is added to the {{{Skipped}} files instead. We will also add tests to the Skipped files if it would affect the rest of the test run or cause NRWT itself to crash.

In the Chromium ports, the convention is to add a line of text to the test_expectations.txt file (see below).

Lastly, we also support the concept of "reference tests", which check that two pages are rendered identically (pixel-by-pixel). As long as the two tests' output match, the tests pass. For more on reference tests, see RefTests.

Suppressing failures using NRWT: the test_expectations.txt file

The test expectations file is found in a platform-specific directory under LayoutTests. I will use the Chromium version as an example.

Syntax

The syntax of the file is roughly one expectation per line. An expectation can apply to either a directory of tests, or a specific tests. Lines prefixed with " " are treated as comments, and blank lines are allowed as well.

The syntax of a line is roughly:

<modifier> <modifier>* ":" <test-name> "=" <expected result>+

For example:

BUGWK12345 WIN DEBUG : fast/html/keygen.html = CRASH

which indicates that the "fast/html/keygen.html" test file is expected to crash when run in the Debug configuration on Windows, and the tracking bug for this crash is bug #12345 in the webkit bug repository. Note that the test will still be run, so that we can notice if it doesn't actually crash.

Expected results

The expected result can be one of PASS, FAIL, TEXT, IMAGE, CRASH, TIMEOUT, IMAGE+TEXT, AUDIO. These should be fairly self explanatory. Note that IMAGE+TEXT means that we expect *both* the text output and the image output to be different. In other words, this is an AND, not an OR.

The "FAIL" modifier, on the other hand, is an OR, and is equivalent to saying that any one of TEXT, IMAGE, AUDIO, or IMAGE+TEXT might happen, but not TIMEOUT or CRASH.

Multiple expected results are allowed, for tests that are flaky and may produce different results in different runs.

Modifiers ===

The set of allowed modifiers are a bit more complicated ... they include bug identifiers, configuration parameters, and miscellaneous options.

Bug identifiers allow you to identify the tracking bugs. They must start with a "BUG" prefix. There are three supported identifiers:

1. BUGWK12345 indicates a bug in the WebKit bug database, and is equivalent to ​https://bugs.webkit.org/show_bug.cgi?id=12345
2. BUGCR12345 indicates a bug in the Chromium bug database, and is equivalent to ​https://bugs.chromium.org/12345 .
3. BUGDPRANKE is a "placeholder" that indicates that no bug has been filed yet, but you should bug that individual about the status.

Configuration parameters describe which variations of your port the test expectation should apply to. Typically each test_expectations.txt file is used for multiple variations of a test run, because each variation usually has a lot of failures in common, and it's easier to manage all of the failures in one place. The exact set of configuration parameters will vary from port to port. Here are the options supported for Chromium:

• LEOPARD SNOWLEOPARD XP VISTA WIN7 LUCID : these indicate particular versions of particular operating systems. Multiple options may appear on a single line, but duplicates are not allowed
  • MAC WIN LINUX: These are "macros" that expand out to all of the relevant versions of each operationg system. It is a syntax error to specify both one of the macros and one of the expanded options, e.g. "MAC LEOPARD"
• DEBUG RELEASE: the different build types
• GPU CPU: Whether we are testing the "GPU-accelerated" code paths, or the regular software-only (CPU) code paths
• X86 X86_64: Whether we are testing the 32-bit or 64-bit versions of the code.

Different ports may not support all of these options, since they may not be relevant.

Note that not all parameters need to be listed, and if no parameters in a particular category are listed, the parser assumes that any combination applies.

To figure out if a configuration matches, we take the logical OR of the modifiers in each category, and the AND of modifiers in different categories, so a line containing "LEOPARD VISTA DEBUG GPU" means that it will apply to tests run on either Mac Leopard or Windows Vista, in Debug mode, using the GPU acceleration code paths.

There are also "miscellaneous" modifiers:

• SLOW : This indicates that the test is expected to be slow, and we apply three times the normal timeout for a test. Note that SLOW tests cannot be expected to TIMEOUT.
• SKIP: This indicates that the test will never pass and there's no point in running it. This is equivalent to listing the test in the Skipped files (see below)
• WONTFIX: This is a modifier that does not affect anything in the test run itself, but can be used for reporting; it indicates that we don't ever expect the test to pass.
• NOW: This modifier has never been used and should probably be removed.
• REBASELINE: This modifier is used by the rebaseline-chromium-webkit-tests script, and is not allowed to exist in a checked-in version of the file.

Semantics

When parsing the file, we use two rules to figure out if an expectation line applies to the current run:

1. If the configuration parameters don't match the configuration of the current run, the expectation is ignored.
2. Expectations that match more of a test name are used before expectations that match less of a test name.

For example, if you had the following lines in your file, and you were running a debug build on Mac SnowLeopard:

BUGWK12345 SNOWLEOPARD : fast/html = TEXT
BUGWK12345 SNOWLEOPARD : fast/html/keygen.html = PASS
BUGWK12345 VISTA : fast/forms/submit.html = IMAGE

You'd expect:

• fast/html/article-element.html to fail with a text diff (since it is in the fast/html directory)
• fast/html/keygen.html to pass (since the exact match on the test name)
• fast/html/submit.html to pass (since the configuration parameters don't apply.

Again, *duplicate expectations are not allowed*.

Suppressing failures using ORWT: Skipped files and checked-in failures

ORWT has a much simpler mechanism. Tests are either expected to pass, or can be skipped. To skip a test, list it in the Skipped file for your platform. If your test produces output different from the "expected" version, check in the new (possibly incorrect) version in your platform-specific directory. See LayoutTestSearchPath for figuring out where that directory is.