Instant feed rendering via two-level cache and lazy hero banner

[cloudflix-x]

Summary

Implemented an intelligent Two-Level Provider & Homepage Cache Architecture (In-Memory ConcurrentHashMap + Persistent Disk SharedPreferences JSON serialization) combined with Lazy Hero Banner Loading to accelerate home screen rendering, eliminate cold-start blank screens, and allow movie lists to populate instantly via a Stale-While-Revalidate pattern.

---

Code Changes & Architecture

1. Two-Level Caching Engine (HomeCache.kt)

• L1 In-Memory Cache: Thread-safe ConcurrentHashMap for sub-millisecond retrieval of active feeds during session navigation.
• L2 Persistent Disk Cache: JSON-serialized disk storage surviving process kills and cold app restarts.
• Data Models: Added CachedHomeData, CachedHomePageResponse, and CachedSearchResponse supporting serialization across all TvTypes (Movie, TvSeries, Anime, Live, Torrent).

2. ViewModel & Lazy Hero Banner Integration (HomeViewModel.kt)

• Instant Feed Hydration: Modified load() to immediately dispatch cached data via HomeCache.getHomeCache(api.name), rendering movie lists directly from local disk cache.
• Decoupled Lazy Hero Banner: Moved hero banner trailer and metadata fetching into an asynchronous background coroutine (addJob = ioSafe { ... }), preventing network calls from blocking initial feed appearance.
• Background Revalidation: Saves fresh network data to L1 & L2 cache on getMainPage() success without causing UI flicker or focus resets.

3. API Load Response Caching (APIRepository.kt)

• Extended API load response caching to respect user-configured TTL (cacheTimeSeconds) and enabled/disabled state.

4. User Preferences & Storage Management (SettingsProviders.kt & DataStoreHelper.kt)

• Cache Duration Setting: Added user-configurable TTL (Off / Always Reload, 10m, 30m, 1h, 3h, 6h, 1d, 3d, 1w) persisted in DataStoreHelper.cacheTimeMinutes.
• Clear Provider Cache Action: Added one-tap cache invalidation across home and provider caches with dynamic disk usage calculation (formatShortFileSize).

5. Automated Unit Tests (HomeCacheTest.kt)

• Comprehensive test suite covering JSON serialization/deserialization, TTL expiration, memory eviction, multi-threaded access, and cache clearance.

---

Performance Impact (Xiaomi TV - Real Data)

Test Environment: Xiaomi TV (MiTV-MZTU0 / Android TV 14, Quad-Core Cortex-A55, Mali GPU, 2 GB RAM, 1080p@60Hz UI)
Target Provider: StreamPlay (20 Home Sections / Movie Lists)
Benchmark Flow: Cold Start ➔ Time to First Movie List (TTFML) ➔ Home Feed Render (via adb logcat, dumpsys gfxinfo, and dumpsys meminfo)

Metric	Without Cache (master)	With Two-Level Cache + Lazy Banner (feat/home-provider-cache)	Improvement
Time to First Movie List (TTFML)	~70,000 ms (70s) (Full network & JSON parsing)	2,348 ms (2.3s) (Direct from L2 Disk Cache)	-96.6% (67.6s faster)
Hero Banner Hydration	Blocking main flow	8,050 ms (Lazy in Background)	Instant list, zero UI blocking
Startup Screen State	Blank / Loading spinner	20 Movie sections rendered immediately	Instant 10-foot interaction
Total Memory PSS	175.24 MB (Total RSS: 281.01 MB)	124.37 MB (Total RSS: 234.71 MB)	-29.0% (~50.9 MB saved)
Java Heap Allocated	50.37 MB	34.06 MB	-32.4% Lower heap allocation
D-pad Frame Pacing (P50)	31 ms	27 ms	Smoother scrolling on TV CPU

---

Validation

• ./gradlew testPrereleaseDebugUnitTest (All HomeCacheTest cases passed)
• Verified on real Xiaomi TV device (MiTV-MZTU0 / Android TV 14) via ADB instrumentation

AI Disclosure

• AI Usage: ~45% of the cache serialization boilerplate, ViewModel wiring, and unit test cases were generated/assisted by AI.
• Verification: The code was verified via automated unit tests and profiled on physical Android TV hardware.

Screenshots

Before (Without Cache / Eager Network Load)	After (Two-Level Cache + Lazy Hero Banner)

Provider Cache Settings & Duration Selector

[fire-light42]

Good idea, and great performance testing but this pull request is way too bloated, with cache logic interspersed with normal code. Surely you can move the cache logic to APIRepository to reduce the complexity and boilerplate by a lot.

This whole pull request (excluding tests) ought to be achievable in ~200 lines, not 700.

[cloudflix-x]

@fire-light42 My bad for the messy diff before — had some messy merge/cherry-pick conflicts from my dev branch that pulled in unrelated stuff. It's clean and minimal now.