Prior to this commit, the term "harness" was used to refer both to the
like-named directory in the repository and the external projects which
consume the tests. Reserve the term "harness" to refer solely to the
files maintained within this project's `harness/` directory, and use
the term "runner" to refer to the external projects.
Prior to this commit, the term "helper" was not defined and therefore
introduced ambiguity in text that described harness files. Replace the
term with more explicit wording in terms of the harness files.
Because test runners have some leeway in how they include harness files,
the requirement that "[t]he test source code must not be modified in any
way" does not necessarily preclude the use of harness files. Update the
instructions for interpreting the `raw` frontmatter flag to explicitly
disallow harness files.
This change is in service of forthcoming tests for the "JSON modules"
language proposal [1]. Verifying the semantics of that proposal requires
modules whose source text is not valid ECMAScript; this change updates
the guidelines for contributing and interpreting tests so that such test
material can be handled consistently.
Differentiating JSON files with a distinct file name suffice will assist
consumers which require special handling of such files (e.g. web
browsers).
Change the pattern used to designate "fixture" files so that it may be
applied to files used for JSON modules.
Increment the project version number to alert consumers of this change
in interpreting instructions.
[1] https://github.com/tc39/proposal-json-modules
Prior to this commit, the implementation of Markdown on GitHub.com
caused the contents of two list items to be rendered outside of their
containing list. Update the indentation so that the structure of the
rendered output matches the structure implied by the text.
The following tests call `IsHTMLDDA` with more than one argument:
- annexB/built-ins/String/prototype/split/custom-splitter-emulates-undefined.js
- annexB/built-ins/String/prototype/replaceAll/custom-replacer-emulates-undefined
- annexB/built-ins/String/prototype/replace/custom-replacer-emulates-undefined.js
Migrating all tests to this API is necessary to prevent tests from hanging indefinitely when a SAB is sent to a worker but the code in the worker attempts to create a non-sharable TypedArray (something that is not Int32Array or BigInt64Array). When that scenario occurs, an exception is thrown and the agent worker can no longer communicate with any other threads that control the SAB. If the main thread happens to be spinning in the $262.agent.waitUntil() while loop, it will never meet its termination condition and the test will hang indefinitely.
Because we've defined $262.agent.broadcast(SAB) in https://github.com/tc39/test262/blob/master/INTERPRETING.md, there are host implementations that assume compatibility, which must be maintained.
Also fixes multiple issues and bugs in various Atomics tests.
CONTRIBUTING.md and INTERPRETING.md:
- Add missing description for `CanBlockIsFalse` flag and also introduce `CanBlockIsTrue` flag which is needed for some tests (see below).
harness/atomicsHelper.js:
- Updated `$262.agent.waitUntil` to clarify it can also be called with BigInt64Array objects.
- Added `$262.agent.timeouts` constants to unify the various, inconsistently used timeouts in the Atomics tests. Each timeout constant has a brief description and a usage example to clarify when it should or should not be used. The default values were tested in SpiderMonkey (locally and CI) under various system load levels and should hopefully also be valid for other engines.
- Added `$262.agent.tryYield` to yield control from the main thread. The default implementation simply calls `$262.agent.sleep`, but test262 hosts could theoretically provide a different implementation.
- Added `$262.agent.trySleep` to replace direct calls to `$262.agent.sleep` from the main thread. Motivation for this function: Some test262 hosts (like browsers) may be able to pause and sleep on the main thread, so they could provide their implementation for `trySleep` which performs a busy-wait or something do nothing.
harness/testAtomics.js:
- Replace `let`, `arrow-functions`, and `for-of` loops to avoid creating extremely long `features` lists in tests using this helper. Removed `Map`, `WeakMap`, `Set`, and `WeakSet` (these weren't even listed in features.yaml) for the same reason and added the missing `Symbol` entry for this file to features.yaml.
- Updated all files including "harness/testAtomics.js" to match the new feature requirements.
test/built-ins/Atomics/store/good-views.js and test/built-ins/Atomics/store/bigint/good-views.js:
- Replace `arrow-functions` and `for-of` loops with ES5 alternatives.
test/built-ins/Atomics/wait/bigint/*.js and test/built-ins/Atomics/wake/bigint/*.js
- Some tests were using `BigInt64Array.BYTES_PER_ELEMENT * 8` for the SharedArrayBuffer length, but their non-BigInt counterparts are using `Int32Array.BYTES_PER_ELEMENT * 4`. For consistency and to make it easier to compare the BigInt against the non-BigInt versions, I've changed it to `BigInt64Array.BYTES_PER_ELEMENT * 4`.
- Also aligned formatting and statement placement when they differed between the non-BigInt and the BigInt version of a test file. (I've diffed some of the non-BigInt and BigInt files against each other and different formatting was a nuisance.)
Test files using `$262.agent.monotonicNow()`:
- Moved `$262.agent.report()` calls outside of the block of code measured `$262.agent.monotonicNow()` to avoid measuring how long it takes to execute `$262.agent.report()`.
- Without this change some tests failed intermittently in certain test configurations in SpiderMonkey. For example with the flags `--ion-eager -- ion-offthread-compile=off` which forces early Ion compilation on the main thread. The `$262.agent.report()` implementation in the SpiderMonkey test262 host embedding uses a for-loop which was forcefully Ion compiled under these settings. And because Ion compilation can take some time, the test case ran longer than `$262.agent.MAX_TIME_EPSILON` which lead to intermittent failures.
Test files using `CanBlockIsFalse` / `CanBlockIsTrue`:
- Some of these tests actually expected that the main thread can wait and [[CanBlock]] is `true` for the agent record executing the test. Therefore I've added a new `CanBlockIsTrue` flag and replaced the flags where needed.
test/built-ins/Atomics/wait/**/*.js and test/built-ins/Atomics/wake/**/*.js:
- Use an atomic counter `RUNNING` in more tests to have better control when a worker agent was actually started.
- Replace the various `$262.agent.sleep(/* Sleep X ms to ensure worker actually sleeps */)` calls with the new `$262.agent.tryYield()` function. This `X` was sometimes as low as 10 milliseconds, which is definitely too short for CI systems under heavy load (observed by intermittent CI failures for SpiderMonkey) and sometimes as high as 500 milliseconds, which is probably much longer than needed even when the system is under heavy load.
- Removed duplicate strings in assertion messages, presumably from copy-pasting the messages between different files.
- Removed extra empty lines at the end of multiple files.
test/built-ins/Atomics/wake/bad-range.js, test/built-ins/Atomics/wake/bigint/bad-range.js, and test/built-ins/Atomics/wait/bigint/non-bigint64-typedarray-throws.js:
- Removed unnecessary `features` and `includes` from this file.
test/built-ins/Atomics/wait/waiterlist-order-of-operations-is-fifo.js and test/built-ins/Atomics/wait/bigint/waiterlist-order-of-operations-is-fifo.js:
- The test was actually broken and didn't test what it said it does. This probably explains #1530.
- The test wants to ensure the waiterlist is implemented as a FIFO structure. This requires that the waiting agents all wait on the same index position, because the waiterlists are defined by each index. But if the agents wait on different indices, each agent is inserted into a different FIFO structure and therefore we can't observe any FIFO ordering between the agents when they're woken up.
- All this requires a bit of synchronization between the main agent and the waiting agent, I hope the added comments help to review these changes.
test/built-ins/Atomics/wait/good-views.js:
- The agent sends multiple reports to the main agent, but only the first one was read.
- This error was introduced during previous refactorings. I've changed it to back to use a while-loop as used in the first version of this file.
test/built-ins/Atomics/wait/no-spurious-wakeup-no-operation.js:
- Only the BigInt version of this test was present, copied it so we also get code coverage for the non-BigInt case.
test/built-ins/Atomics/wait/waiterlist-block-indexedposition-wake.js:
- Added extra while loops to avoid intermittent failures when the agent worker haven't started to wait.
- This should help to avoid some of the intermittent failures we saw for SpiderMonkey.
test/built-ins/Atomics/wake/wake-all-on-loc.js and test/built-ins/Atomics/wake/bigint/wake-all-on-loc.js:
- This test was also no longer after previous refactoring sessions.
- The "B" agent only waited for 10 milliseconds, which made it likely that it already timed out before the main agent was able to call `Atomics.wake`, which in turn rendered the test useless, because the test case wants to ensure that `Atomics.wake` cannot wake "B". But if "B" was already timed out, it can trivially not be woken by `Atomics.wake`.
- Added some safety measure to catch the case when "B" timed out before `Atomics.wake` was called and made it a test error if that happens.
test/built-ins/Atomics/wake/count-defaults-to-infinity-missing.js, test/built-ins/Atomics/wake/count-defaults-to-infinity-undefined.js, and test/built-ins/Atomics/wake/undefined-index-defaults-to-zero.js:
- Changed the `$262.agent.start()` calls to use a for-loop to avoid code duplication.
- (Forgot to undo the code formatting around `assert.sameValue`, still need to change it back.)
- Also more while-loops around `Atomics.wake`.
- These changes should fix#1529 and #1566.
test/built-ins/Atomics/wake/wake-all.js:
- Removed "B" worker agent.
- Without this removal the test case would be exactly equal to test/built-ins/Atomics/wake/wake-all-on-loc.js.
test/built-ins/Atomics/wake/wake-in-order-one-time.js:
- Add for-loops to avoid code duplication and make the test more readable.
- Make the `Atomics.wake` assertion messages unique by adding the current loop counter.
- Add `$262.agent.tryYield()` to give the worker agents enough time to actually start waiting.
test/built-ins/Atomics/wake/wake-in-order.js:
- Removed the outer loop `attempt < 10` because it uses `$262.agent` in a way currently not required to work. And which actually also doesn't work in SpiderMonkey's implementation of `$262.agent`.
- According to INTERPRETING.md `$262.agent.broadcast()` broadcasts its message to all agents ever started by `$262.agent.start()` and then blocks until all agents received the broadcast. It is not required that the agents started by the first `$262.agent.broadcast()` call will all be disabled/destroyed/whatever when the second `$262.agent.broadcast()` call occurs, which then means the second `$262.agent.broadcast()` call still tries to reach the agents started in the first loop iteration, but these may no longer accept broadcasts and therefore won't acknowledge they've received the broadcast. Which then means the second `$262.agent.broadcast()` call waits forever.
Previously, the error message generated by failed asynchronous tests was
generic and underspecified. Improve the format and explicitly document
it in project's interpreting guidelines.
Early errors may result from parsing the source text of a test file, but
they may also result from parsing some other source text as referenced
through the ES2015 module syntax. The latter form of early error is not
necessarily detectable by ECMAScript parsers, however. Because of this,
the label "early" is not sufficiently precise for all Test262 consumers
to correctly interpret all tests.
Update the "phase" name of "early" to "parse" for all those negative
tests that describe errors resulting from parsing of the file's source
text directly. A forthcoming commit will update the remaining tests to
use a "phase" name that is more specific to module resolution.
Modify $262.uncallableAndIsHTMLDDA() to an optional $262.IsHTMLDDA (whose use must be guarded by a feature of the same name), and narrowly/correctly prescribe its requirements consistent with `document.all`'s behavior in HTML.
Changes to the instructions for interpreting tests will likely produce
new failures for consumers who are updating between revisions of
Test262. Introduce a machine-readable convention for signaling
substantive changes.
Previously, test consumers were encouraged to insert a `throw` statement
as the first statement of tests for early errors. This recommendation
made tests harder to consume, and as an optional transformation,
consumers may have ignored it or simply been unaware it was made. By
explicitly including such a `throw` statement, the tests become more
literal, making them easier to consume and more transparent in their
expectations.
Document expectation for all tests for early errors to include an
explicit `throw` statement. Extend linting script to verify that
contributors are automatically notified of violations and to ensure that
future contributions satisfy this expectation.
Define the expected behavior of new host-defined utilities. These will
facilitate forthcoming tests that concern cross-realm and cross-script
semantics (which are currently untestable using standard ECMAScript
alone).
The project's CONTRIBUTING.md was written with test authors in mind. It
contains details on non-technical metadata (e.g. "author" and "es6id"),
helper function usage, and preferred code structure. In addition, it
elides certain low-level technical details on the requirements for the
runtime environment.
Introduce a new document targeted towards those executing the tests.
Formalize all expectations regarding how the runtime environment should
be defined, how metadata should be interpreted, and how results should
be understood. This information has overlap with the CONTRIBUTING.md
file, but it also contains details that are irrelevant to test authors.
This document can serve as a more formal contract between Test262 and
the implementors who consume it. This allows Test262 to unambiguously
document future modifications to the formal requirements which in turn
supports consumers who maintain their own test harnesses.
Chengzhong Wu
Philip Chimento
Mike Pennisi
Michael Dyck
Rick Waldron
Leo Balter
André Bargull
Claude Pache
Ross Kirsling
Mathias Bynens
mitchcurtis
Jeff Walden
Thomas Wood
Shu-yu Guo