test: stop two tests failing on timing and megapixels

[thymikee]

Two tests blocked unrelated PRs today. Neither is auto-retryable: #1419's contention retry declines find.test.ts as outside its enumerated list, and slow-gate failures cannot be re-checked by a rerun. Every occurrence costs someone a manual re-run.

find.test.ts — an un-assertable poll count

wait captures fresh snapshots while polling asserted exactly 2 dispatches.

The wait loop polls at a 300 ms interval against a 350 ms budget, and its final sleep consumes whatever remains — so it lands on exactly remainingMs() === 0. A sleep that returns even a millisecond early makes hasTimeRemaining() true and admits a third poll. The count was never assertable; it just usually came out at 2.

The test's actual subject is that each poll re-captures rather than reusing the first tree. It now asserts that directly: at least two captures, and every one a snapshot.

The sibling exact-count assertion earlier in the file is deliberately left alone — it takes its count from mockResolvedValueOnce ordering (the match appears in the second tree), not from the clock, so it is deterministic.

apps.test.ts — 3.2 megapixels to prove a ratio

Two tests built 1206×2622 source PNGs (iPhone 16 Pro native) purely to verify rescale arithmetic. Both are now 126×273 — still divisible by 3, so the /3 and 2/3 rescales stay exact — with expectations updated to match.

Measured locally:

test	before	after
keeps requested simulator pixel density	1021 ms	34 ms
retries simulator capture timeouts	1176 ms	233 ms
file test time	2.48 s	1.00 s

Verification

18/18 find.test.ts, 54/54 apps.test.ts. Typecheck, lint and oxfmt clean.

Neither change weakens what the tests cover: the polling test still proves fresh captures, and both rescale tests still assert exact output dimensions.

---

Generated by Claude Code

[github-actions]

Size Report

Metric	Base	Current	Diff
JS raw	1.91 MB	1.91 MB	+72 B
JS gzip	613.3 kB	613.3 kB	+88 B
npm tarball	731.3 kB	731.0 kB	-328 B
npm unpacked	2.57 MB	2.56 MB	-2.9 kB

Startup median (7 runs, lower is better):

Scenario	Base	Current	Diff
CLI --version	27.4 ms	26.1 ms	-1.3 ms
CLI --help	57.7 ms	56.4 ms	-1.3 ms

Top changed chunks:

Chunk	Raw diff	Gzip diff
dist/src/ime-lifecycle.js	+512 B	+234 B
dist/src/session.js	0 B	+2 B
dist/src/internal/daemon.js	+2 B	+1 B

[thymikee]

Clean review at exact head efc7a746e388ba1d1a0f9725121b10c58620bbcc. The find timeout regression now asserts the actual contract—at least two polls and every poll dispatching a fresh snapshot—without pinning scheduler-dependent timing. The screenshot fixtures retain exact scale-3 arithmetic (42×91 and 84×182) plus retry/count coverage while avoiding multi-megapixel allocations. All exact-head checks and platform smokes are green; no live-device evidence is needed for this deterministic test-only change. Ready for human merge.

Residual risk: no separately authorized cross-vendor review was performed.

[github-actions]

PR Preview Action v1.8.1
Preview removed because the pull request was closed.
2026-07-31 14:16 UTC