[thymikee]

Summary

Use synthesized coordinate taps for resolved iOS in-app back controls. Retain measured candidate frames through ranking and dispatch to avoid repeated element resolution and XCTest tap quiescence waits. Keep the navigation-bar → ranked back control → verified leading-slot fallback and #2728 typed outcomes. macOS/tvOS behavior is unchanged. Three files; no scope expansion. Related: #2728.

Validation

Base: 642393cadb4510cdccab581c1046ac83e9b52ae8. Head: 0351d1996998593599a9a24806298747f2a545c0.

• pnpm check:affected --run --head 0351d19969: exit 0, exactCommit=true; receipt .tmp/validation/run-YyADUC/receipt.json.
• pnpm format, pnpm check:xctest-selection, TypeScript build, iOS runner build: passed.
• Two focused native tests: head 2 passed; same tests/fixture with base navigation: 2 failed on the intended XCTest-tap assertion. Independent review: no findings.
• Bluesky 1.136.0, dedicated iPhone 17 Pro/iOS 27.0: identical post-thread → back → feed sequence, three successful runs each. Runner accepted→completed milliseconds: base 2091/2378/2331; head 1397/1461/1527. CLI wall milliseconds: base 2382/2705/2535; head 1588/1720/1860. Median runner time improves 37.3%; all six navigations verified on the feed.

CI still owns macOS/native cross-platform builds and integration/coverage. Review control-selection parity and the existing synthesis-failure fallback. No PR was opened by this worker.

Implemented by a Codex worker; opened and reviewed by the coordinating Claude session.

🤖 Generated with Claude Code

[github-actions]

Size Report

Metric	Base	Current	Diff
Installed (including dependencies)	5.52 MB	5.52 MB	-6.7 kB
Package (unpacked)	5.52 MB	5.52 MB	-6.7 kB
Package (download)	1.67 MB	1.67 MB	-2.7 kB

Startup median (7 runs, lower is better):

Scenario	Base	Current	Diff
CLI --version	44.9 ms	43.9 ms	-1.0 ms
CLI --help	89.5 ms	92.2 ms	+2.6 ms

[thymikee]

Independent review (Codex, read-only) of 0351d19969 raised one P2, fixed in ed1a21f98a.

P2: the navigation-bar branch replaced back.tap() with a frame-centre synthesized tap. isHittable promises a hittable point, not a hittable centre, so a bar button partly under an overlay could pass selection while its centre missed. That branch now keeps XCTest's element tap, exactly as on main. The ranked back control (the branch Bluesky's "Go back" takes, which already tapped its centre) keeps the synthesized dispatch, so the measured gain is unchanged. The navigation-bar test now asserts exactly one element tap.

Validation at ed1a21f98a:

• pnpm check:affected --run --head: passed.
• testRankedInAppBackUsesSynthesisWithoutXCTestTap and testNavigationBarBackKeepsTheElementTap: passed on an iOS 27.0 iPhone 17 Pro simulator.

[thymikee]

This PR is ready at ed1a21f. All 16 checks pass on that commit, and I know of no conflicts. Not blocking, and you can take or leave it: tapResolvedNavigationBackControl in RunnerTests+Navigation.swift repeats the same synthesis-then-XCTest-fallback tap that tapTopLeadingNavigationFallback uses in the same file, so one small private helper that taps a point and reports whether it dispatched could serve both branches. The Bluesky timing ran on 0351d19, not ed1a21f, so one back run on ed1a21f whose runner log shows the coordinateTap synthesis marker would confirm that the ranked branch carries the gain. Nothing else needs to happen before merge.

[github-actions]

PR Preview Action v1.8.1
Preview removed because the pull request was closed.
2026-10-22 02:55 UTC