Optimize API calls and enhance concurrency in TraktRepository - #122

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, you can upgrade your account or add credits to your account and enable them for code reviews in your settings.

The concurrency work in this PR is genuinely valuable — moving IptvRepository off manual thread pools onto coroutine-based parallelism, the coroutine-safe TMDB season cache in TraktRepository, parallel stale-record cleanup in TraktSyncService, and the new TmdbTvSeason model are all good improvements. CI is green and the diff is mergeable.

However, this PR cannot be merged as-is because it also adds a large number of developer scratchpad / AI-assistant artifacts at the repo root that should not ship with the app. Please remove the following before merge:

One-off patch scripts (all at repo root):

• fix_brace.py
• fix_brace2.py
• fix_calc.py
• fix_callers.py
• fix_concurrency.py
• fix_tier2.py
• patch_atomic.py
• patch_auth.py
• patch_benchmark.py
• patch_cancellation.py
• patch_comment.py
• patch_distinct.py
• patch_iptv.py
• patch_mutex.py
• patch_requestjson.py
• patch_requestjson2.py
• patch_requestjson3.py
• patch_timeout.py
• update_iptv_repo.patch

Misplaced Kotlin file:

• IptvBenchmark.kt at the repo root — this is not under app/src/..., so it isn't part of the module source set and shouldn't be committed. The actual benchmark test already lives at app/src/test/kotlin/com/arflix/tv/data/repository/IptvBenchmarkTest.kt, which is the correct location.

Quick cleanup:

\\�ash
git rm fix_.py patch_.py update_iptv_repo.patch IptvBenchmark.kt
git commit --amend --no-edit # or a new commit
git push --force-with-lease
\\

After that cleanup the remaining diff (~6 production files + 2 tests) is the real change and is worth a careful re-review before merge, since IptvRepository is core to playback. Specifically I'd like to verify:

1. The suspend conversion of requestJson and the coroutine cancellation semantics don't introduce new cases where a cancelled caller leaves an OkHttp call in flight.
2. The new limited-parallelism dispatcher bounds are sane for the typical EPG fetch volume.
3. The new TraktRepository TMDB-season cache has a bounded lifetime or invalidation path (to avoid stale season metadata after content updates).

Happy to give it a full review once the scratchpad files are gone.

Thanks for addressing every concern from the earlier review. Re-audited the final diff against current main (which now includes the 14 bug/feature PRs that merged this morning) and everything checks out:

Blockers resolved:

• All 19 scratchpad fix_*.py / patch_*.py / update_iptv_repo.patch files removed in 21f02d4.
• Root-level IptvBenchmark.kt removed (the real benchmark test lives at app/src/test/kotlin/.../IptvBenchmarkTest.kt as expected).
• Final diff is exactly the 7-file production + test change I said would be merge-worthy after cleanup.

Three original concerns verified:

1. requestJson suspend conversion cancellation semantics — textbook correct. continuation.invokeOnCancellation { call.cancel() } cancels the in-flight OkHttp call, if (continuation.isActive) guards prevent double-resume, response.close() handles the arrived-after-cancellation case. No in-flight leaks.

2. Limited-parallelism dispatcher bounds — Dispatchers.IO.limitedParallelism(20) with 60s withTimeoutOrNull is a direct 1-for-1 replacement for the old newFixedThreadPool(20) + 60s awaitTermination. The small-favorites path went from min(10, size) to a fixed 20, but that's still bounded and safe.

3. TMDB-season cache lifetime — my concern assumed a long-lived @Singleton field; the author correctly chose an ephemeral ConcurrentHashMap<Pair<Int,Int>, Deferred<TmdbSeasonDetails?>> that's a local variable of each batch enrichment call. Allocated per enrichContinueWatchingItems invocation, populates during awaitAll, garbage-collected when the batch finishes. No stale-data possible because each new home load creates a fresh cache. The putIfAbsent + CompletableDeferred pattern is the standard atomic single-flight implementation.

Additional fixes from the review-comment iterations that I also verified:

• supabaseAuthMutex.withLock { authRepository.refreshAccessToken() } in TraktSyncService — serializes concurrent token refreshes and prevents conflicting DataStore writes. Genuine bug fix, not just a theoretical concern.
• if (e is CancellationException) throw e in the parallel stale-playback cleanup — correctly preserves structured concurrency semantics.
• inFlightRequests guard in AnimeMapper.ensureSeasonEpisodeCountsCached — per-show single-flight dedupe with explicit cleanup in both success and exception paths (inFlightRequests.remove(tmdbId)), so the map doesn't grow unbounded.
• Nullability fix in calculateTmdbSeasonOffset return type (Int → Int?) — callers now correctly distinguish "couldn't determine offset" from "offset is 0".

Minor nits (not blockers, not addressing them):

• IptvBenchmarkTest uses Thread.sleep() inside async { ... } instead of delay(), so it's actually measuring thread blocking rather than coroutine semantics. The benchmark numbers it prints aren't a fair comparison, but since the test only prints to stdout and asserts nothing, it can't regress or flake anything.
• TraktRepositoryBenchmarkTest asserts 1 == 1 as a placeholder — the author's comment honestly notes that properly mocking TraktRepository requires Robolectric or MockK which aren't set up in this project. Zero test value but zero harm.
• Pre-existing catch (e: Exception) in AnimeMapper that swallows CancellationException without rethrow — inherited from original code, not introduced by this PR, so not a regression. Worth fixing in a follow-up.

Mergeability: CLEAN. Dry-run merge against current main confirms no conflicts. My 14 merged PRs' added files (ApkInstallReceiver.kt, JikanApi.kt, MobileBackButton.kt, etc.) all remain untouched because the PR branch doesn't reference them.

Squash-merging now. Thanks for the thorough follow-through — this is a genuinely valuable concurrency cleanup that lays the groundwork for richer TV metadata (the new TmdbTvSeason/seasons field).