Note: tests are not the only part of the codebase that uses `std`. Other
parts, like `tools/`, do too. So, it could be argued that this is a
little misleading. Either way, I'm doing this as discussed with
@mmastrac.
Running `tools/wpt.ts` with a filter would cause an error if there were
extra, leftover expectations in expectations.json. These errors would
not appear if no filter was passed, often leaving the filtered version
of the test runner broken.
This also introduces a smarter bit of logic where filters can be
specified with a leading slash (`tools/wpt.ts run -- /url` is equivalent
to `tools/wpt.ts run -- url`)
Fixes the WPT tests that test w/invalid codes. Also explicitly ignoring
some h2 tests to hopefully prevent flakes.
The previous changes to WebSocketStream introduced a bug where the close
errors were not made available if the `pull` method was re-entrant.
This speeds up WPT tests in two ways:
1. The `WebCryptoAPI` tests are slow, so create a parallel bucket for
each individual test instead of one for all the `WebCryptoAPI` tests.
2. If a test is expected to fail, use a shorter timeout (1 minute rather
than 4).
To make it easier to debug which tests are slowing us down.
Tests taking more than 5s have duration printed in red,
taking more than 1s in yellow and less than 1s are printed
without color.
- relands #17872
- updates the timeouts to be re-configurable just for CI
- fixes `./tools/wpt.ts update`
- adds option not "ignore" during, applied to wpt epoch runs only
Updated third_party dlint to v0.37.0 for GitHub Actions. This PR
includes following changes:
* fix(prefer-primordials): Stop using array pattern assignments
* fix(prefer-primordials): Stop using global intrinsics except for
`SharedArrayBuffer`
* feat(guard-for-in): Apply new guard-for-in rule
* 'request-upload.h2' and 'redirect-upload.h2' only work with a
functional HTTP2 test harness server, otherwise they're flaky.
* Fetch request streaming tests require a server that doesn't choke
on requests that use 'Transfer-Encoding: chunked'.
Classic workers were implemented in denoland#11338, which also enabled the WPT
tests in the `workers` directory. However, the rest of WPT worker tests
were not enabled because a number of them were hanging due to
web-platform-tests/wpt#29777. Now that that WPT issue is fixed, the bulk
of worker tests can be enabled.
There are still a few tests that hang, and so haven't been enabled. In
particular:
- The following tests seem to hang because a promise fails to resolve.
We can detect such cases in non-worker tests because the process will
exit without calling the WPT completion callback, but in worker tests
the worker message ops will keep the event loop running. This will be
fixed when we add timeouts to WPT tests (denoland#9460).
- `/fetch/api/basic/error-after-response.any.worker.html`
- `/html/webappapis/microtask-queuing/queue-microtask-exceptions.any.worker.html`
- `/webmessaging/message-channels/worker-post-after-close.any.worker.html`
- `/webmessaging/message-channels/worker.any.worker.html`
- `/websockets/Create-on-worker-shutdown.any.worker.html`
- The following tests apparently hang because a promise rejection is
never handled, which will kill the process in the main thread but not
in workers (denoland#12221).
- `/streams/readable-streams/async-iterator.any.worker.html`
- `/workers/interfaces/WorkerUtils/importScripts/report-error-setTimeout-cross-origin.sub.any.worker.html`
- `/workers/interfaces/WorkerUtils/importScripts/report-error-setTimeout-redirect-to-cross-origin.sub.any.worker.html`
- `/workers/interfaces/WorkerUtils/importScripts/report-error-setTimeout-same-origin.sub.any.worker.html`
This commit implements classic workers, but only when the `--enable-testing-features-do-not-use` flag is provided. This change is not user facing. Classic workers are used extensively in WPT tests. The classic workers do not support loading from disk, and do not support TypeScript.
Co-authored-by: Luca Casonato <hello@lcas.dev>
Currently, a WPT test is considered failed if its status code is
anything other than 0, regardless of whether the test suite completed
running or not, and any subtests that haven't finished running are not
considered to be failures.
But a test can exit with a zero status code before it has completed
running, if the event loop has run out of tasks because of a bug in one
of the ops, leading to false positives. This change fixes that.
These reports can be consumed by tools like `wptreport` or
https://wpt.fyi. The old style report could be removed in a future PR
when wpt.deno.land is updated.
