feat(booth): photo-booth strip — 5-4-3-2-1 ×3 capture into a printed strip

[stozo04]

Description

Implements docs/PRD-photo-booth.md — the classic photo-booth artifact: a provisional Booth button runs a self-driving 5-4-3-2-1 countdown ×3 with a flash flicker per shot, top-bias square-crops each viewfinder grab at grab time, and composites the three frames into one vertical strip (white borders, "OPENLOOP · AUG 20 2026" footer, color by default with a B&W toggle). The strip saves through the unmodified photo path: savePhoto → MediaStore Pictures/OpenLoop → share sheet → in-app gallery. The lens tray stays interactive during every countdown — the swap window is the product hook.

Key design points (all per PRD sign-off):

• No CaptureMode entry, no new OpenLoopUiState — booth is ephemeral UI state in CameraScreen; the exhaustive router when is untouched (Lesson 014). Zero camera-binding changes (Lesson 031 never comes into play).
• BoothStripLayout (pure, Android-free, JVM-tested) owns all geometry — including the D3 regression test that a geometric center crop of a 9:19.5 source is rejected (it clips the crown / above-eye lens art). BoothStripComposer is a thin Canvas pass; B&W is a zero-saturation ColorMatrixColorFilter on the frame draws only, so borders/footer stay crisp either way (D4).
• captureBoothStrip: injectable composer seam (JVM tests stay on the TestDispatcher — Lesson 029), re-entrancy guard mirroring photoSaveInProgress, and a §5.4 guard — an incomplete frame set (null grab abort) surfaces the existing capture-failed snackbar and never saves a partial strip.
• Accessibility (in the never-simplify-away bucket): countdown digits + "Shot n of 3" are polite live regions; Booth, Cancel, and the B&W chip carry content descriptions; the chip has switch semantics.
• Cancel / predictive back abort the sequence and discard frames (Lesson 015); shutter, mode selector, and camera flip are disabled mid-sequence; the B&W chip stays live mid-countdown via rememberUpdatedState (Lesson 034).

Related Issue

Closes #135

Type of Change

• ✨ Feature: A new feature (non-breaking change which adds functionality)

How Has This Been Tested?

• Automated: :app:testDebugUnitTest — 125 tests in the touched suites, 13 new (BoothStripLayoutTest ×7, OpenLoopViewModelTest booth ×6), 0 failures. :app:connectedDebugAndroidTest on Pixel 8 (API 36 AVD) — 121 tests, 0 failures incl. 3 new booth UI tests. Debug and release builds green; :app:lintDebug — 0 findings in any booth file (one KTX warning found and fixed pre-commit); 16 KB zipalign -c -P 16 real pass (every .so (OK) at 16384-multiples).
• Manual (Pixel 8 emulator, full flow driven end-to-end): Booth tap → 5-4-3-2-1 ×3 with flash → share sheet opens on the composited strip → "Saved" → strip in the in-app gallery and Pictures/OpenLoop. Second run with B&W toggled → grayscale frames, crisp borders/footer. Cancel mid-sequence → idle restored, file count unchanged. Countdown digits/Cancel/tray never appear in the captured frames.
• Static analysis: Android Lint clean (Engine 1). Engine 2 "Inspect Code" not run — headless inspection is non-functional on this machine (see docs/STATIC_ANALYSIS.md fallback); substituted lint + zero compiler warnings.

Screenshots

What could not be verified + manual QA checklist (needs a real face / device)

• Elvis pompadour / broccoli crown stays inside each square on a tall selfie (the crop math is JVM-tested; the visual needs a face — emulator's virtual scene has none)
• Swapping lenses during a countdown bakes different lenses into different frames, in color (success criterion 6)
• TalkBack pass: countdown ticks and "Shot n of 3" announced politely; Booth/Cancel/B&W labels read correctly
• Known accepted behaviors (all documented in §5.1/§5.4 as of 0fe00ef): home (gallery) button mid-sequence silently aborts; backgrounding (ON_STOP) mid-sequence silently aborts — after started-then-stopped the TextureView serves a frozen frame, so grabs can't be trusted (Lesson 036); activity recreation mid-sequence resets to idle; after the third grab the strip is committed (hand-over precedes the cosmetic final flash, ebbdf87) — Cancel/gallery/back/backgrounding during that 250 ms can no longer discard it
• TalkBack: shutter announces disabled mid-booth (real enabled, not a no-op) — needs a TalkBack pass to hear it
• Backgrounding abort on a real device: take a phone call mid-countdown → sequence discards, no strip, no share sheet (emulator equivalent verified by code path only)
• Samsung/Exynos reverse on media3 1.11.0: the upgrade is emulator-verified (reverse.complete, frames=9 sync=9); the Samsung encoder try-order lanes (Lessons 023/027) need the RTL/OEM sweep or a real Galaxy

Checklist

• 🧪 My changes have been verified locally and work as expected.
• 🔍 I have performed a self-review of my own code.
• ✍️ I have commented my code, particularly in hard-to-understand areas.
• 📖 My changes generate no new compile warnings or errors.
• 🧹 The git branch is clean and references to obsolete branches have been pruned.

🤖 Generated with Claude Code

https://claude.ai/code/session_014W2oDvqCgazqBaBjJ2kJWC

Review follow-up (2026-08-20)

Two commits after the /code-review high pass + Cursor Bugbot review:

• 0fe00ef — all 10 review findings + both Cursor comments closed. Lifecycle-gated grabs (backgrounding = Cancel, Lesson 036 added), genuinely disabled shutter mid-booth, flash/overlay state reset at sequence start, captureBoothStrip guard reorder (abort snackbar can't be swallowed; completed strip survives the gallery race; frames owned + recycled on every path; OutOfMemoryError on the composite degrades to the snackbar), one shared savePhotoAndShare tail + one shared save guard, glassPill() dedupe, crop off-main, and a real-CameraScreen instrumented test.
• ebbdf87 — Cursor round 2 (gallery tap drops finished strip): valid, fixed. The hand-over now happens at the third grab, before the final flash — the dominant disposal race (effect cancelled at animateTo before captureBoothStrip ran) is gone by construction, and the point of no return matches D4's "until the last grab" boundary. PRD + KDoc updated; README refreshed (compileSdk 37, booth feature, 536/122 counts).
• 487d338 — all 25 pre-existing lint warnings cleared. AGP 9.3.1 · Kotlin 2.4.10 · compileSdk 37 · compose BOM 2026.08.00 · media3 1.11.0 · core-ktx/lifecycle/fragment/work/test/uiautomator/firebase/plugins to current stable; UseKtx fixed; OldTargetApi suppressed at the source with the reason (targetSdk 36 is the deliberate, tracked decision — Play floor is 35). :app:lintDebug now reports an empty issue list.

Post-change gate: clean assembleDebug + assembleRelease green (exit 0, zero e:, zero w: after fixing the 10 compile warnings a clean build surfaced) · 16 KB zipalign -c -P 16 real (OK) on every .so · 536 unit tests, 0 failures · 122 instrumented tests, 0 failures (Pixel_8 AVD — note the image is now API 37) · app driven live: booth 5-4-3-2-1 ×3 → strip → share sheet, and a 3.2 s FORWARD_THEN_REVERSE boomerang save (reverse.complete | frames=9 sync=9, zero FATAL in logcat). Engine 2 "Inspect Code" not run (headless inspection non-functional on this machine — STATIC_ANALYSIS.md fallback; substituted lint + zero compiler warnings).

---

Note

Medium Risk
Touches camera capture, bitmap memory, and the shared photo-save path, plus a Media3/AGP/Kotlin/compileSdk upgrade. targetSdk stays 36, but reverse/export and OEM lanes should be re-checked after the stack bump.

Overview
Adds a photo-booth strip: arm Photo Booth in the lens drawer, then the shutter runs a 5-4-3-2-1 countdown ×3, square-crops each viewfinder grab (top-biased so crown/lens hair stay in frame), and composites a vertical JPEG (white borders, OpenLoop + date footer, color or B&W). The strip reuses the existing photo save/share path — no new CaptureMode or camera bind.

Booth is ephemeral UI on CameraScreen. Mid-sequence the shutter/mode/flip are disabled, lenses stay live for swaps, and cancel/back abort until the last grab (then the strip is committed before the cosmetic flash). Backgrounding aborts because PreviewView.getBitmap() can return a frozen frame after ON_STOP (Lesson 036). Incomplete sets, overlapping saves, and OOM on composite all fail to the existing capture-failed snackbar and recycle bitmaps.

Also bumps compileSdk 37, AGP 9.3.1, Kotlin 2.4.10, Compose BOM, Media3 1.11.0, and related libs; targetSdk stays 36 on purpose. Version 1.0.44. JVM layout tests plus ViewModel/UI coverage for the booth flow.

^{Reviewed by Cursor Bugbot for commit 44f4e47. Bugbot is set up for automated code reviews on this repo. Configure here.}

[stozo04]

PR Review — Google Android Standards Compliance

Reviewer: Claude (Automated)
Date: 2026-08-20
PR: #138 — feat(booth): photo-booth strip — 5-4-3-2-1 ×3 capture into a printed strip
Standards sourced from: Target API level requirements · Compose performance best practices · Compose phases · Compose semantics / accessibility · Merging & clearing semantics · Coroutines best practices · CameraX architecture · UI layer / UDF · Testing strategies · Request runtime permissions · DataStore · Android 16 behavior changes
Files reviewed: 18 changed (2 new Kotlin sources, 2 modified Kotlin sources, 3 test files, strings, CLAUDE.md, PRD, cspell.json, 6 proof images) plus the untouched files booth depends on (MainActivity.kt event handling, VideoStorageRepositoryImpl.savePhoto, CameraManager)

---

PASS

• Architecture Booth bypasses CaptureMode and adds no OpenLoopUiState entry — the exhaustive router when is untouched, exactly as PRD-photo-booth §5.1 and Lesson 014 require. captureBoothStrip follows UDF: UI hands frames up, state/events flow down (OpenLoopViewModel.kt). The composer enters via constructor injection (composeBoothStrip seam) mirroring the logged publishPhotoToLibrary pattern; the ViewModel stays Context-free (a Bitmap is plain data — Lesson 004).
• Compose Countdown state reads are deferred into BoothCountdownOverlay's own scope via digit: () -> Int / shot: () -> Int lambdas, and the flash alpha is read inside graphicsLayer { } (draw phase) — a 1 Hz tick or a flash frame never recomposes the AndroidView viewfinder tree (defer reads). Effect-captured values are aliased through rememberUpdatedState (currentBoothMonochrome), closing the Lesson 034 trap for the mid-countdown B&W toggle — and the chip is actually reachable mid-sequence (rendered beside Cancel), so the alias does real work.
• CameraX Zero camera-binding changes: no new use case, no rebind, no CameraManager edit. The grab is PreviewView.getBitmap() per the shipped PRD-photo-capture D1 decision. Camera flip is disabled mid-sequence, so no rebind can blank the ritual. Lessons 012/022/031 invariants hold (grep "CameraScreen(" MainActivity.kt still returns 1).
• Coroutines All work in viewModelScope; the composite runs on Dispatchers.Default behind an injected suspending seam (tests substitute a TestDispatcher-bound fake — no real-dispatcher hop for virtual time to miss, Lesson 029). CancellationException is rethrown before the best-effort catch (OpenLoopViewModel.captureBoothStrip). The one broad catch (Exception) is the MediaStore publish, a documented deliberate deviation carried over from capturePhoto (PRD-photo-capture §5.5).
• Media & Storage The strip rides the existing photo path wholesale — savePhoto → JPEG in filesDir/videos/ (inside the file_paths.xml share boundary) → best-effort MediaStore publish → share sheet with ClipData (Lesson 028, unchanged code). Bitmap memory is bounded: crop-at-grab retains ~1080² per frame, frames are recycled after composite, the strip after save, and cropToBoothSquare guards the documented createBitmap aliasing case (Bitmap docs) before recycling the source.
• Testing 13 new JVM tests + 3 new instrumented tests, all green (125 unit / 121 connected total, 0 failures). Highlights: BoothStripLayoutTest locks the D3 regression (a center crop of a 9:19.5 source is explicitly rejected); the re-entrancy test parks savePhoto on a gate so overlap is enforced, not hoped for (Lesson 029); the B&W chip test drives the toggle both ways to catch a frozen lambda (Lesson 034). Pure math lives in an Android-free file per the repo's extract-the-math pattern.
• Accessibility All three new controls are ≥48dp, carry content descriptions, and use semantic roles (Role.Button, Role.Switch on the toggleable chip). The countdown digits and "Shot n of 3" are LiveRegionMode.Polite live regions, so TalkBack announces the ticks and shot advances — a timed visual-only countdown would have excluded non-sighted users (Compose semantics). The digit node stays mounted through the grab beat so the live region persists.
• Permissions No new permissions; booth runs entirely inside the already-granted camera session. Degradation is graceful: a null grab aborts to the existing "Couldn't take that photo" snackbar (MainActivity.kt:382), never a crash or a partial strip.
• DataStore Untouched; the deliberate non-persistence of the B&W choice is logged in the PRD (out of scope) rather than an omission.
• Play Store / Android 16 The API 36 floor is now in force for new apps/updates as of August 31, 2026 (requirements, confirmed today) — OpenLoop already targets 36, so this PR ships inside the window. Predictive back is handled: the booth-active flag joins the gated BackHandler and back aborts through UI state, per the target-36 rules (ANDROID_STANDARDS §11, Lesson 015).

FAIL

None.

WARNING

None introduced by this PR. (Pre-existing lint OldTargetApi and dependency-freshness warnings predate the PR — see Static Analysis below.)

RECOMMENDATIONS

• [i18n] boothFooterText uppercases the formatted month with Locale.getDefault() — media/BoothStripLayout.kt. In Turkish (dotted/dotless I) and some non-Latin locales the uppercased month can render oddly next to the English "OPENLOOP" wordmark.
  • Why: a printed artifact with mixed-script casing looks like a bug to those users; pinning the date format to the wordmark's locale (or accepting and documenting default-locale dates) makes it deliberate either way.
  • Effort: low.
• [Compose/UX] The B&W chip fires no haptic on toggle while its sibling CaptureModeSelector performs ToggleOn/ToggleOff (ui/CameraScreen.kt, issue Camera/video mode toggle gives no feedback, and the icon is hidden under your finger while tapping #126 rationale: the tap should confirm itself through touch).
  • Why: consistent feedback across the two adjacent toggle surfaces; trivially testable with the existing recording-haptics pattern in CameraScreenTest.
  • Effort: low.
• [Timing] The countdown accumulates delay(1_000L) ticks, so the ~18 s ritual stretches on CPU-starved devices (the same drift Lesson 012's hand-off note measured on the recording timer). No correctness impact — nothing is capped by it — but if the booth cadence ever needs to be wall-clock-honest, inject a monotonic time source as that note prescribes.
  • Why: documented repo-known drift; cheap to fix later because the loop is one constant.
  • Effort: low (when needed).
• [Static analysis noise, pre-existing] 25 lint warnings on main carry into this branch unchanged: 20× GradleDependency, 2× AndroidGradlePluginVersion, 1× NewerVersionAvailable (expected dependency-freshness background per docs/STATIC_ANALYSIS.md), 1× OldTargetApi (the app targets 36, the current floor — noise until a 37 target exists), and 1× UseKtx at camera/lens/LensSurfaceProcessor.kt:625. None are from this PR; the one UseKtx the PR did introduce was caught and fixed before merge (BoothStripComposer.kt now uses the KTX createBitmap).
  • Why: keeping the "every finding is live" no-baseline policy honest.
  • Effort: n/a for this PR.

---

Summary

Category	Pass	Fail	Warning	Rec
Architecture	1	0	0	0
DataStore	1	0	0	0
Permissions	1	0	0	0
Compose	1	0	0	1
CameraX	1	0	0	0
Media & Audio	1	0	0	0
Coroutines	1	0	0	1
Testing	1	0	0	0
Accessibility	1	0	0	1
Play Store	1	0	0	0
Android Version	1	0	0	0
Static Analysis (Lint + IDE Inspect)	1	0	0	1
Total	12	0	0	4

Static analysis engines: Engine 1 (Android Lint) ran — :app:lintDebug clean of PR-introduced findings (0 in any booth file; the 25 carried warnings are itemized above). Engine 2 (IDE "Inspect Code") was NOT run — headless inspection is non-functional on this machine (docs/STATIC_ANALYSIS.md fallback applies); it must be run locally in Android Studio before merge, or its absence accepted explicitly. Tier 3 OSS fallback ran on the changed Markdown: markdownlint, cspell, and link-check all pass after in-PR fixes (fence language + list spacing in the PRD; nine legitimate jargon terms added to cspell.json).

Verdict

APPROVE

This PR does what the signed-off PRD says and nothing it doesn't: the strip geometry is pure and regression-tested against the exact crop mistake the PRD forbids, the capture sequence adds zero camera-binding risk, the save path is reused rather than re-implemented, and the accessibility work (live-region countdown, labeled controls, switch semantics) is real rather than decorative. Verification is unusually complete — both build variants, 246 green tests, a genuine 16 KB zipalign pass, and an emulator-driven end-to-end run with the produced JPEGs inspected (color, B&W, and a mid-countdown toggle). The one thing to do before merge: run Engine 2 (IDE Inspect Code) locally per docs/STATIC_ANALYSIS.md, and keep the PR's own manual-QA checklist for the face-dependent checks (lens hair inside the square, lens-swap-per-frame) on a real device — the emulator cannot exercise either.

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit b7e7f2e. Configure here.}

[stozo04]

UI redesign — booth moves into the lens drawer (D2 placement decided)

Commit ef7abbf lands the owner-approved placement from the "Photo Booth Controls" design canvas (Option E, iterated live 2026-08-20):

• Idle viewfinder is clean — the provisional floating Booth + B&W pills are gone.
• Two-tab drawer behind the bottom-left button: Photo Booth | Lenses slider (Lenses default). The Photo Booth tab swaps the carousel for the D4 Color | Black & White radio pair and arms the booth; the shutter starts the strip while armed (announced as "Start photo booth").
• Armed state survives closing the drawer (lime lens button = the cue); picking Camera/Video disarms; the booth tab's ✕ cancels a running sequence + disarms + closes (predictive back still aborts). No separate mid-sequence Cancel/B&W row — the drawer is the control surface.
• D5 banner during the sequence: "Swap lenses between shots", lime-trimmed glass chip, center-aligned with the gallery button.

Verification (Definition of Done)

Gate	Result
:app:assembleDebug / :app:assembleRelease	BUILD SUCCESSFUL, exit 0, zero e:
:app:testDebugUnitTest	536 tests, 0 failures
:app:connectedDebugAndroidTest (Pixel_8 AVD, API 37)	121 tests, 0 failures, 1 pre-existing skip — full suite re-run after the final UI tweak. One earlier run hit an empty <failure/> on SpeedTabPanelCurveTest (instrumentation process death under load, no assertion message); it passed 15/15 in isolation and the full re-run was green
:app:lintDebug	No issues found (0 errors, 0 warnings)
Engine 2 (IDE Inspect Code)	Not run — cannot execute headlessly on this machine (per docs/STATIC_ANALYSIS.md fallback: lint + compiler-warning-free build substituted)
16 KB zipalign	Not re-run — this commit is pure Kotlin/resources/docs; no packaging or native-lib change
Run on emulator + screenshots	App driven through onboarding → camera → drawer → arm → countdown; screenshots captured of idle, Lenses tab, armed Photo Booth tab, and the running countdown with banner (owner also exercised the flow live on hardware during the iteration)

Manual QA still worth a device pass

• Full 3-shot strip on hardware with a face: arm from drawer → strip saves via the unchanged photo path, Color and B&W both
• Lens swap mid-countdown from the drawer's Lenses tab bakes different lenses into different frames (D5)
• Back-gesture priority mid-sequence with the drawer open (closes drawer first, then cancels)

🤖 Generated with Claude Code

https://claude.ai/code/session_01Wkq4T1GP4jcvxkitbxeQAS