Repository navigation
Tags
Tags
Wait for the player's report before stopping the capture under test Summary: `AudioCaptureSessionTests.stopMixesEverythingAConnectedProcessPlayedIncludingWhatItHadLeft` failed intermittently: the captured process came back with no name (`captured.processes.map(\.name) → [""]`) instead of `["Player"]`. The test stopped the capture as soon as the scripted engine began. A process is named by the report its shim client sends once it has written audio, on the client's first poll, so the capture could stop before that report arrived. The session then never read the report. It still mixed the process's audio, but without a name, which is how it treats any process whose report it has not read when the capture stops. The test now stops the capture only once the client has sent its report. The scripted engine signals on its second drain: the client sends what the first drain produced before it polls again, so by then the report is waiting for the session. Product code is unchanged. ___ landed-with-radar-review Differential Revision: D128935104 fbshipit-source-id: 61d472e6bdf4541bc18545a80e11e232015cc886
Add the AVFoundation objects a virtual camera hands to apps Summary: Add Objective-C subclasses for the AVFoundation device, format, input, connection, discovery and photo objects supplied by the virtual camera. Their initialisers install the private state required by AVFoundation deallocation; the init-family replacements live in Objective-C for the same reason. Provide native-call result/exception helpers for the Swift camera layer and build this as the ARC static library `ShimulatorVideoInjectionAVFoundation`. It is not yet linked into `libShimulator-iOS`. Reviewed By: mark-cmd-alt Differential Revision: D124174605 fbshipit-source-id: ec4812efa012c488310d4fbc3c05ed1715cea3a4
Cover the persistent transport not retrying a cancelled round trip Summary: No behavioral change. The persistent axbridge transport retries a read once when its round trip fails, but not when the round trip was cancelled: its caller has gone. That branch had no test; a read whose connection reports `CancellationError` now asserts the failure is `CancellationError` and the request was sent once. Differential Revision: D125677437 fbshipit-source-id: 9ccf071344800cc9932be59652f51a317a0e0957
Use AsyncEvent for the tests' hand-rolled gates Summary: Two `FBSimulatorControl` test files used their own gate actors, each holding a list of continuations it released once, to sequence their concurrency tests: `AsyncGate` in the install preflight tests and `Gate` in the HID operation tests. They now use `FBControlCore`'s `AsyncEvent<Void>`, whose `waitIgnoringCancellation()` keeps the gates' non-cancellable waits. Both actors are deleted. The tests assert the same orderings as before. Differential Revision: D125655011 fbshipit-source-id: f1f01778fa8780c067fc1bfae8511d8b426f3bb1
Stop the idb-companion formula test expecting libRepl-iOS.dylib Summary: `brew test idb-companion` passes again on a companion built since the REPL moved into libShimulator. The formula's test asserted `Resources/libRepl-iOS.dylib`, which builds no longer ship, so it failed with `Expected path '…/libexec/Resources/libRepl-iOS.dylib' to exist`. The test now checks the Resources a current companion ships: libShimulator, both SimulatorFrameworkBridges, ReplHost.app and IDBAPI.swiftinterface. The template and the committed v1.7.3 formula change together, so the tap stays the render of its recorded release; `idb-release.json` records the new formula's digest. v1.7.3's own tarball still contains libRepl-iOS.dylib, and dropping the assertion doesn't change whether the formula installs and tests correctly for that release. Differential Revision: D125653825 fbshipit-source-id: 7c6963d29b1bc5b5bf73036a58d907e07c652dec
Stop bounding the unanswered accessibility read test by wall-clock time Summary: `AXReadTimeTests.anUnansweredElementStopsTheReadAfterOneTimeout` no longer fails when the host is slow. It required the read to take under 100 ms, to show that only one of its 50 ms requests was waited out. A loaded CI runner took longer than that to wait out the one request. The test already shows "one, not two" by its request count (the root, then the first child, which went unanswered), so it keeps that and the lower bound, which shows the request's timeout was waited out, and drops the upper bound. No production code changes. This is a timing race in the test itself, so it can't be pinned by a deterministic test first; the evidence is the CI failure below. Differential Revision: D125467056 fbshipit-source-id: 3e954550ea87356c85f3359537b93b9f73132fe6
Wait for accessibility through the companion's current no-element wor… …dings Summary: The end-to-end harness again waits while the simulator's accessibility service is not ready yet, with companions that name the read that found no element. Before, it recognized only the older "No translation object returned", so with a current companion: - the accessibility probe before the suite failed setup on the first "The simulator's accessibility service named no frontmost application", failing every end-to-end test; - a read that found no element was not repeated; - the accessibility tests' wait for the fixture app's row failed instead of relaunching the fixture and waiting. Each of these now accepts the older wording and the four current ones (a point, the frontmost application, a pid, and an element that is no longer there). Writes are still never repeated on any of them. Differential Revision: D125248266 fbshipit-source-id: 84f27e5396e51d16c41deca13e0758efb803bccd
Move UIAutomationTests to Swift Testing Summary: No behavioral change. `UIAutomationTests` now runs under Swift Testing instead of XCTest: the `XCTestCase` subclass becomes a `Suite`, each `test…` method an `Test` function without the `test` prefix, and the `XCTAssert…` calls become `#expect` and `#require`. What each test checks is unchanged. Differential Revision: D124864491 fbshipit-source-id: 3fc0099d9eae9cba32164a06651a46378c9f9b90
Skip deploying the site for a CI run that was cancelled Summary: The website deploy workflow no longer fails when the CI run that triggered it was cancelled. A push to `main` cancels the CI run still going for the previous push. That run can stop before its `Build the website from the end-to-end run` job exists, and `demos-built` then failed with `CI run <id> has no job named 'Build the website from the end-to-end run'`, a red run on `main` for nothing wrong. A cancelled run built no site, so `demos-built`, and with it `deploy-website`, is now skipped for one. A completed run without that job still fails, which keeps the guard against renaming the job in `ci.yml`. This can't be pinned by a unit test: it's a GitHub Actions `if` on the triggering run's conclusion, which only a real `workflow_run` exercises. Differential Revision: D124942961 fbshipit-source-id: dcac348882d3f51d264a127674a15367a935af34
Give client types stable names Summary: No behaviour change. `BridgeAXClientType` gains an explicit `name` and `init?(name:)`, one spelled-out string per case. The `complete` document's `client.type` and sime2e's `--client-type` both use it, where they used `String(describing:)` of the case before. The names are an output and command-line contract. Deriving them from Swift case labels meant renaming a case would silently change what a read reports and which flag values a caller can pass. Every name equals the case label it replaces, so output and accepted flags are unchanged. Differential Revision: D124702210 fbshipit-source-id: e2c4efe61b4b50f5a70db03c986b9c73949d43ba
PreviousNext