revert: carry one phase Deadline through the runner interfaces (#2473)

[thymikee]

Reverts #2473 (f4c8f3ddda) to restore the iOS live lanes on main.

Evidence

• main iOS: green at 3e89f821d6 (16:00), red at f4c8f3ddda (16:44, first run containing refactor(apple): carry one phase Deadline through the runner interfaces and test cancellation as a matrix #2473) on Run iOS Settings replay smoke test (replay step 6 wait timed out after 3 attempts).
• refactor(apple): carry one phase Deadline through the runner interfaces and test cancellation as a matrix #2473's own post-rebase run at 16:15 failed Run fixture-backed iOS simulator E2E smoke (wait text "Agent Device Tester" 10000 timed out). chore(layering): runner modules reach host-kit only through the runner host port #2470 and chore(gates): name the added modules and import paths when an eager closure grows #2471, run at the same minute on the same base without refactor(apple): carry one phase Deadline through the runner interfaces and test cancellation as a matrix #2473, passed.
• chore(gates): name the added modules and import paths when an eager closure grows #2471 rebased onto f4c8f3ddda fails the same fixture smoke at 17:00; fix/ios-record-video-host-busy-2170 fails at 16:52.
• Session log of the failing wait (artifacts of run 34503978665 / chore(gates): name the added modules and import paths when an eager closure grows #2471 run at 17:00): open finished in 4.3 s; the wait's ios.snapshot-source.prepare reported request canceled at 5.2 s, the runner fallback logged ios_runner_session_reuse, ios_runner_readiness_preflight, ios_runner_connect ×2 and retry ×3, and the wait failed at 10.4 s. A later plain snapshot in the same session succeeded (6.2 s), so the runner itself was alive.

#2473 changed only packages/platform-apple/src/runner/* (the startup, adoption, reuse and build phases now spend one RunnerPhaseBudget). The live reproduction is not yet isolated to a line; reverting first, fixing forward with a reproduction on a booted simulator second.

No test changes are lost that matter to main: the cancellation matrix and the phase-budget interface come back with the fix-forward PR.

[github-actions]

Size Report

Metric	Base	Current	Diff
Installed (including dependencies)	4.54 MB	4.54 MB	+194 B
Package (unpacked)	4.54 MB	4.54 MB	+194 B
Package (download)	1.35 MB	1.35 MB	+47 B

Startup median (7 runs, lower is better):

Scenario	Base	Current	Diff
CLI --version	27.7 ms	27.7 ms	+0.0 ms
CLI --help	76.4 ms	75.3 ms	-1.1 ms

[thymikee]

Ready at 4353e76. This restores the exact tree before #2473, and the iOS simulator replay and smoke lanes now pass along with the other checks. Please keep the phase-budget refactor out until a fix-forward change reproduces and resolves the live failure.

[thymikee]

Withdrawing this revert. A local reproduction on a booted simulator did not flip between f4c8f3ddda (#2473) and its parent fea7ca8a43: the CI settings replay (test … 01-settings.ad, --retries 0) passed 4/4 on both, and an instrumented open → click → wait produced byte-identical request logs on both, served by a warm AX bridge in ~107 ms.

The CI evidence, re-read against the artifacts, does not support #2473 either:

• Every failing wait shows the same shape: ios_runner_session_reuse → connect ok in 7–59 ms → ios_runner_readiness_preflight (1 s) → a second ios_runner_connect that stalls 4.5–5.1 s → the wait's own 5 s / 10 s deadline cancels it (captures: 1, readableCaptures: 0). Every budget number in those logs (preflight 1000 ms, connect ~45 000 ms, command ~44 9xx ms) is byte-identical to a green run without refactor(apple): carry one phase Deadline through the runner interfaces and test cancellation as a matrix #2473, and the green run shows the same second connect at 11 ms to 2.3 s, i.e. the same distribution inside the margin. The difference is runner latency on a cold host, not a budget or cancellation semantic.
• Two of the four "correlated" red runs (fix/replay-deep-link-classifier-b, fix/ios-record-video-host-busy-2170) failed a pure Swift XCTest step on a different image (Xcode 26.6 / iphonesimulator 26.5) with ALERT_DEADLINE_EXCEEDED; they cannot involve refactor(apple): carry one phase Deadline through the runner interfaces and test cancellation as a matrix #2473 and I should not have counted them.
• refactor(apple): carry one phase Deadline through the runner interfaces and test cancellation as a matrix #2473's own run failed the settings replay on attempt 1 and passed on attempt 2. This revert's green iOS run is one sample.

Code reading found no defect: the only semantic delta in #2473 is requireRunnerPhaseRemainingMs losing a numeric fallback that was unreachable at all three call sites.

The real problem is the readiness preflight abandoning a runner command that keeps the runner busy for ~5 s; filed separately with the artifact evidence. Closing without merging; #2473 stays on main.

[github-actions]

PR Preview Action v1.8.1
Preview removed because the pull request was closed.
2026-09-10 17:49 UTC