Repository navigation
Fragments between key frames - #3449
Closed
robspychala wants to merge 571 commits into
Closed
robspychala wants to merge 571 commits into
robspychala wants to merge 571 commits into
Conversation
…selection PiperOrigin-RevId: 975569196
Returning an incomplete ListenableFuture from onGetLibraryRoot() or onConnectAsync() previously deadlocked the service main thread because Futures.addCallback posted the completion callback to the application looper while onGetRoot() was synchronously blocking that same looper. Pre-resolve controller parameters and command buttons up-front, and invoke the completion callback using MoreExecutors.directExecutor() to unblock the main thread immediately upon future completion. Issue: androidx#3393 #cherrypick PiperOrigin-RevId: 975693221
See: https://developer.android.com/jetpack/androidx/releases/compose-ui#1.13.0-alpha01 that brought minSdk up to 24 which we now pull into the demo-effect via material3-alpha26 dependency. PiperOrigin-RevId: 975709854
PiperOrigin-RevId: 975762775
This change ensures nextTrackId is consistently initialized to 1 and uses track.id directly when writing tfra boxes and processing tracks, preventing off-by-one track ID discrepancies in multi-track fragmented MP4 containers. PiperOrigin-RevId: 975869862
Reverts `MiniController` to use the standard `bitmapLoader` overload rather than the custom artwork slot. When sharing an artwork composable between the `ContentFrame` and the `MiniController`, using `fillMaxSize` was interfering with the sizing that `MiniControllerTokens.ArtworkSize` tries to enforce. PiperOrigin-RevId: 975958834
PiperOrigin-RevId: 975969937
ClippingSlider(progress, clippingRange, ...) overload that became unused when ClippingSlider switched to `RangeSliderState`. Add a TODO explaining that once `onValueChange` is supported by `RangeSliderState`, the `minRangeDelta` clamping logic can be moved to that callback instead of observing snapshotFlow. PiperOrigin-RevId: 976019967
This adds support for 32-bit floating-point PCM audio tracks in Mp4Muxer and FragmentedMp4Muxer by writing the standard fpcm sample entry and pcmC configuration box per ISO/IEC 23003-5. Issue: androidx#3389 PiperOrigin-RevId: 976225331
On devices with strict EGL drivers (such as Qualcomm Adreno 740 on Samsung Galaxy Z Flip 5), attempting to create an EGL window surface on a destroyed or invalid Surface throws an IllegalArgumentException. Adding a surface.isValid() check before surface creation prevents this crash and treats the debug preview render as a safe no-op. PiperOrigin-RevId: 976229076
This change appends outgoing MediaQueueItem payloads with unique synthetic IDs, to uniquely identify them when returned back by the Cast receivers. PiperOrigin-RevId: 976276156
Material3-alpha25 brought the following: - Update `rememberModalBottomSheetState` to include `rememberBottomSheetState` with `initialValue = SheetValue.Hidden`. - Update `ListItem` usages to use the trailing lambda syntax for `headlineContent`. PiperOrigin-RevId: 976287011
Introduced the general-purpose, decoupled PlaybackState (and rememberPlaybackState helper) to reactively expose core player states (isBuffering, isEnded, hasError, isPlaying). Implemented BufferingIndicator Composable which renders an circular spinner. PiperOrigin-RevId: 976291702
Storing a Job in mutableStateOf causes recompositin every time hideJob.value is updated. Since scheduleHideControls() assigns a new Job on every pointer move event (which fires continuously at the screen refresh rate), this will cause severe recomposition thrashing and degrade performance. To prevent unnecessary recompositions, hold the Job in a plain mutable reference inside remember that isn't backed by Compose state. PiperOrigin-RevId: 977407604
In MediaRouteButtonContainer, mediaRouteSelectorListener is registered with cast via registerListenerAndGetCurrentSelector(), but it is never unregistered when the composable leaves composition. If the composable is removed from composition before Cast is initialized, mediaRouteSelectorListener remains in Cast.pendingMediaRouteSelectorListeners, causing a listener memory leak and potentially triggering callbacks on a disposed state. PiperOrigin-RevId: 977418685
* Add an optional field `location` in `DataSpec`, and set the `serviceLocation` on the `DataSpec` for DASH media chunk loading. * Add a package-private `ServiceLocationFilteringThroughputEstimator` within DASH module, which: * Receives transfer events as `TransferListener` from the chunk download `DataSource`; * Assigns individual estimator per `serviceLocation`, each estimator calculates estimates based on a sliding window weighted average of the throughput samples; * Provides throughputs estimate for a `serviceLocation` via `getThroughputEstimate(serviceLocation)`. PiperOrigin-RevId: 977423616
Updates the playback position thumb height to be 1.1 times the slider track height. To prevent the thumb's height from being clamped by the parent constraints, the slider modifier is updated to use unbounded `wrapContentHeight` instead of `fillMaxSize`. PiperOrigin-RevId: 977424770
PiperOrigin-RevId: 977451848
This adds an @ignore test documenting an issue where frames with truncated microsecond timestamps produced by demuxers (such as standard MP4 extractors scaling sample timestamps with RoundingMode.DOWN) are discarded as stale by FrameAggregator when retiming against a virtual clock rounded to the nearest microsecond (RoundingMode.HALF_UP). PiperOrigin-RevId: 977509920
PiperOrigin-RevId: 977512392
This change adds GPS altitude support to Mp4LocationData and updates BoxParser to extract latitude, longitude, and optional altitude metadata from xyz atoms. PiperOrigin-RevId: 977530327
Removes stale TODOs for the following reasons: * Position thumb's height is fixed. * RTL locales: Looking more into this, media-related components should remain in LTR. More info [here](https://m3.material.io/foundations/layout/bidirectionality-rtl#7f9f392d-40d6-4186-86b6-9e59626f963e:~:text=Copy%20link-,Media%20players,-Media%20controls%20for). PiperOrigin-RevId: 977530732
This change adds GPS altitude support to Mp4Muxer, allowing the muxer to write 3-component ISO-6709 coordinate strings (+lat+long+alt/) in the udta xyz box when Mp4LocationData includes an altitude. PiperOrigin-RevId: 977552154
Update documentation for `ClippingSlider`, `ClippingRangeSlider`, and `EditingPlayerScreen` to clarify that the preview bitmap list length must remain constant to avoid layout resizing issues. PiperOrigin-RevId: 977566808
Dynamic scheduling tests in ExoPlayerScrubbingTest previously relied on MediaCodecVideoRenderer and asynchronous decoding via FakeClock, leading to timing races in Robolectric when running under high concurrency. This change switches to a deterministic ForwardingDurationToProgressRenderer that directly returns a fixed durationToProgressUs, matching the test pattern used in ExoPlayerTest. PiperOrigin-RevId: 977887947
Document backwards compatibility and interoperability behavior across MediaSession, MediaController, MediaBrowser, MediaLibraryService, and MediaSessionService when interacting with platform or legacy controllers and sessions. PiperOrigin-RevId: 977911614
Unlike HLS, DASH allows definitions of service location (pathway) for individual elements (eg. Representation). Within the library, the selection of service locations for manifest update is maintained in `DashMediaSource`, while the selection logic for baseUrls for segment fetching is well maintained in `BaseUrlExclusionList`. Instead of pursuing the unity between HLS and DASH `ContentSteeringTracker` implementations as much as possible, it's better to respect their difference, while keeping a small set of shared logics in `BaseContentSteeringTracker`. In `DashContentSteeringTracker`: * We directly propagate the `serviceLocationsPriority` (pathwayPriority) to its callback, and the `DashMediaSource` and `BaseUrlExclusionList` will use it for their selection logics. * We require the owner of the tracker (`DashMediaSource`) to update the `DashManifest` to the tracker, so that the tracker can always track the up-to-date available service locations (this will be useful for pathway clone). * We require a set of `SteeringQueryParameterProvider`s to give a snapshot of selected service locations for any `Location`s or `BaseUrl`s managed under them. PiperOrigin-RevId: 977929176
Introduce InFlightFrameManager to encapsulate tracking and releasing of in-flight Frames. Migrate CompositionVideoPacketReleaseControl and PacketConsumerVideoSampleExporter to delegate in-flight frame management to InFlightFrameManager. PiperOrigin-RevId: 978422178
Conflating **range dragging** with **general user interaction** caused progress scrubbing (`onProgressChange` setting `state.isUserInteracting = true`) to anchor `activeValueRange` to `preDragClippingRange`. On trimmed media, this abruptly shifted the progress slider's bounds and `valueRange` mid-touch, canceling Compose's drag gesture detector and restricting interaction to tap-to-seek only. Additionally, `preDragClippingRange` was not updated when a clipping drag completed. This change: * Introduces a dedicated `isClipping` flag in `ClippingSliderState` to track when clipping handles are actively being dragged. * Anchors `activeValueRange` to `preDragClippingRange` only while `isClipping` is true, ensuring the layout remains fixed during progress scrubs. * Updates `preDragClippingRange = clippingRange` and resets `isClipping` in `onValueChangeFinished`. PiperOrigin-RevId: 978488962
Align backwards compatibility behavior with methods like setMediaButtonPreferences: - Unconditionally update the platform session on setSessionActivity, and ensure standalone sessions without an active media notification controller correctly update and recover the platform session on setPlaybackException. - Forward sessionExtras, sessionActivity, and active playback exceptions configured on AcceptedResultBuilder when the media notification controller connects. PiperOrigin-RevId: 978514749
PiperOrigin-RevId: 992534876
Replace `ListenableFuture.addListener` callbacks and the custom `browser` getter in `MainActivity` and `PlayableFolderActivity` with `kotlinx.coroutines.guava.await()` and `repeatOnLifecycle`, matching `PlayerActivity`. PiperOrigin-RevId: 992550891
Add `MediaMetadata.categories` and `MediaMetadata.Builder.setCategories(List<String>)`. Add `EXTRAS_KEY_SUGGESTED_TYPE`, `EXTRAS_KEY_TTL_SECONDS`, `EXTRAS_KEY_PLAYLIST_TITLE`, `EXTRAS_KEY_PLAYLIST_ARTWORK_URI`, and `EXTRAS_KEY_CATEGORIES` to `MediaConstants`, along with legacy Media1 conversion support. PiperOrigin-RevId: 992626093
When ExoPlayer seeks during audio offload playback, it flushes the audio sink, which triggers `AudioTrackAudioOutput.release()`. Previously, `releaseAudioTrackAsync()` scheduled both `audioTrack.flush()` and `audioTrack.release()` to execute asynchronously after a 20ms delay (`AUDIO_TRACK_VOLUME_RAMP_TIME_MS`). However, when a new `AudioTrack` is immediately created after the seek for the same offload stream, AudioPolicy reuses the same underlying HAL offload stream. If the delayed asynchronous flush from the old track executes after the new track has already started writing data, it clears the HAL buffers that the new track just filled. This causes the new stream to freeze and starve, triggering spurious timestamp mismatch warnings in `AudioTrackPositionTracker`. Flushing before release is only needed to prevent state retention on legacy PCM devices. Offload tracks do not need a flush before release. By skipping `audioTrack.flush()` when releasing offloaded tracks, the shared HAL stream is never flushed by the old track, preventing the race condition while preserving the release delay. PiperOrigin-RevId: 993603427
Pass Period.assetIdentifier to the Period constructor in DashManifest.copy and add unit test coverage in DashManifestTest. PiperOrigin-RevId: 993661238
Using `_` as an identifier has been a reserved keyword since Java 9 (and is only valid for unnamed variables in Java 22+), causing `javac` warnings during Gradle builds targeting Java 11 (`--release 11`). Replace `_` local variable names in `SubripParserTest`, `ProgressiveMediaPeriodTest`, and `AudioCapabilitiesReceiverWrongThreadTest` with `unused`. PiperOrigin-RevId: 993675408
`ExoPlayerMinimalModuleProguard` references `ExoPlayer`, whose compiled class file contains `@RestrictTo(Scope.LIBRARY_GROUP)` annotations. Because `:lib-exoplayer` and `:lib-common` declare `libs.androidx.annotation` as an `implementation` dependency rather than `api`, `androidx.annotation.RestrictTo$Scope` is not exposed transitively on `:test-proguard-minimal`'s compile classpath, causing `javac` to emit `unknown enum constant Scope.LIBRARY_GROUP` warnings when compiling `:test-proguard-minimal`. ``` > Task :test-proguard-minimal:compileDebugJavaWithJavac warning: unknown enum constant Scope.LIBRARY_GROUP reason: class file for androidx.annotation.RestrictTo$Scope not found ``` Add `compileOnly(libs.androidx.annotation)` to `:test-proguard-minimal` so `javac` can resolve class-retention annotations from `:lib-exoplayer` at compile time without adding extra dependencies to the minimal R8 test APK packaging classpath. PiperOrigin-RevId: 993676892
Adds `Flags.FLAG_SKIP_AUDIO_TRACK_FLUSH_BEFORE_RELEASE` (enabled in Canary Mode) to skip calling `AudioTrack.flush()` before `AudioTrack.release()` in `AudioTrackAudioOutput.releaseAudioTrackAsync()` on API 29+. Calling `flush()` before `release()` was introduced in 2015 as a workaround for a legacy passthrough HAL issue where recreated tracks retained the previous playback head position. On API 29+, `AudioFlinger`'s `DirectOutputThread` unconditionally flushes the HAL stream when adding a new track, and skipping the asynchronous flush before release avoids racing with a newly created track sharing the same direct or offload HAL stream. PiperOrigin-RevId: 993680508
Passing `-DHAVE_FSEEKO` on the compiler command line defines the macro as `1` (`#define HAVE_FSEEKO 1`). On 64-bit Android ABIs (`arm64-v8a` and `x86_64`), libflac's `config.h.cmake` (`#cmakedefine HAVE_FSEEKO`) generates `#define HAVE_FSEEKO` with an empty replacement list, causing `-Wmacro-redefined` warnings when `config.h` is included. Using `-DHAVE_FSEEKO=` defines `HAVE_FSEEKO` with an empty replacement list (`#define HAVE_FSEEKO`), which still satisfies `#ifndef HAVE_FSEEKO` in `compat.h` on 32-bit Android ABIs with `minSdk 24` while matching `config.h` identically on 64-bit ABIs. PiperOrigin-RevId: 993684639
Previously, addMediaItems operations were discarded if the receiver had no media items in queue or no operations in flight to add to the queue. This change allows AddMediaItemsOperation to initialize playback when the receiver queue is empty. Issue: androidx#2402 PiperOrigin-RevId: 993704828
When a decoder transiently reports an invalid output format (such as 0x0 on start or flush), onOutputFormatChanged ignores it and leaves pendingVideoSinkInputStreamChange set to true. If the input format queue was already polled, subsequent output buffers may reach processOutputBuffer with pendingVideoSinkInputStreamChange still true. Re-query the codec output format in processOutputBuffer when pendingVideoSinkInputStreamChange is true, and return false if the format is still invalid so the frame is not sent to the VideoSink before onInputStreamChanged is called. This is a follow up on androidx@fa65d31 which didn't fully address the test failure. PiperOrigin-RevId: 993721203
In the Matroska specification (RFC 9559), `ChapterAtom` (`\Segment\Chapters\EditionEntry\+ChapterAtom`) is a recursive master element that can contain nested `ChapterAtom` children. `MatroskaExtractor` previously tracked only a single active `ChapterEntry` field, causing nested `ChapterAtom` elements to overwrite the parent `ChapterEntry` and throw a `NullPointerException` when the parent `ChapterAtom` ended. This change tracks active `ChapterAtom` elements on a stack (`ArrayDeque`) so that nested chapters do not overwrite their enclosing parent chapter. Issue: androidx#3438 PiperOrigin-RevId: 993722199
When `MediaController.setMediaItems` or `prepare` is called while a legacy `MediaSessionCompat` is paused or idle (`!playWhenReady`), `canInitializeLegacyPlaylist` decides whether to eagerly initialize the playlist now via `prepare` / `prepareFrom*` (setting `hasInitializedLegacyPlaylist = true` so a subsequent `play()` only sends `onPlay()`), or wait until `playWhenReady == true` to initialize via `play` / `playFrom*`: - Check `PlaybackStateCompat.getActions()` for the specific prepare action matching the current `MediaItem` (`ACTION_PREPARE` when the item is already known to the session, or `ACTION_PREPARE_FROM_URI`, `ACTION_PREPARE_FROM_SEARCH`, or `ACTION_PREPARE_FROM_MEDIA_ID`) instead of the coarse `Player.COMMAND_PREPARE` flag (which is enabled whenever any single prepare action is supported). - Reset `hasInitializedLegacyPlaylist = false` when the new playlist cannot be eagerly initialized while paused, so a subsequent `play()` call dispatches `onPlayFrom*` instead of a plain `onPlay()`. - In `updateControllerInfo`, check `isLegacyPlaylistInitialized` (`playbackState != Player.STATE_IDLE`, non-empty timeline, and a platform-known `queueId != QueueItem.UNKNOWN_ID`) rather than `canInitializeLegacyPlaylist`, so already-paused platform playlists are marked initialized on connection while pending controller-set items stay uninitialized until prepared or played. PiperOrigin-RevId: 993761019
All methods on `InFlightFrameManager` (`trackIfSuccessful`, `onFrameProcessed`, and `releaseAll`) are confined to a single playback/internal thread per instance: * In `CompositionPlayer` (`CompositionVideoPacketReleaseControl`), `trackIfSuccessful` is only called from `onRender` and `redraw`, `onFrameProcessed` is dispatched via `playbackThreadExecutor`, and `releaseAll` is called from `CompositionPlayerInternal.releaseInternal`, all on `CompositionPlaybackThread`. When `PlayerListener.onPlaybackStateChanged` calls `FrameAggregator.queueEndOfStream` on the application thread, `CompositionVideoPacketReleaseControl.queue` only enqueues packets into `packetQueue`, which is later drained in `onRender` on `CompositionPlaybackThread`. * In `Transformer` (`PacketConsumerVideoSampleExporter`), `HardwareBufferSampleConsumer`'s `frameConsumer` posts to `handlerWrapper`, and `ComponentListener` callbacks (`onWakeup` and `onFrameProcessed`) along with `release` all execute on `Transformer:Internal`. PiperOrigin-RevId: 993784339
- Remove SDK_INT < 24 checks and redundant @RequiresApi(24) annotations - Inline code from nested Api24 wrapper classes PiperOrigin-RevId: 993818116
When container HAGC metadata is not being applied and the stream is high bitdepth AV1 on SDK >= 37, parse AV1 OBUs to extract in-band HAGC metadata and set it explicitly on `MediaCodec` with `setParameters`. This will make sure that we don't rely on each individual MediaCodec implementation to pass this metadata through. Also rewrite all T.35 metadata OBUs as "unknown" so that `MediaCodec` will ignore them. PiperOrigin-RevId: 994034813
This adds support for 24-bit and 32-bit integer PCM audio tracks in Mp4Muxer and FragmentedMp4Muxer by writing the standard ipcm sample entry and pcmC configuration box per ISO/IEC 23003-5. PiperOrigin-RevId: 994150191
Handle `STATE_PROCESS_FULL_INPUT` in `TrimOptimizedExportOperation.getProgress()` and clear `mediaItemInfo` when falling back on `ERROR_CODE_MUXING_APPEND`. Previously, when a trim-optimized export fell back to full input processing, `getProgress()` either remained stuck at `PROGRESS_STATE_WAITING_FOR_AVAILABILITY` (when abandoned before `mediaItemInfo` was assigned) or reported a non-zero starting percentage (when falling back on `ERROR_CODE_MUXING_APPEND` after `mediaItemInfo` was assigned). PiperOrigin-RevId: 994174333
When expanding an X-ASSET-LIST into multiple ads in AdPlaybackState, apply X-PLAYOUT-LIMIT across the expanded ad durations in the AdGroup and mark any trailing ads with zero remaining playout duration as skipped. Keep X-PLAYOUT-LIMIT separate from the interstitial duration in resolveInterstitialDurationUs and only cap the fallback content resume offset by X-PLAYOUT-LIMIT when X-RESUME-OFFSET is omitted and X-PLAYOUT-LIMIT is less than the resolved interstitial duration. When the interstitial duration is unknown (for example before an X-ASSET-LIST JSON response is fetched or when an X-ASSET-URI omits duration attributes), default the fallback content resume offset to 0 even if X-PLAYOUT-LIMIT is set, because X-PLAYOUT-LIMIT is an upper bound on playout time rather than the interstitial duration (RFC 8216bis-22, Appendix D.2). PiperOrigin-RevId: 994272056
`seekBackwards_withDurationLessRawAac_doesNotAdjustSeek` and `seekBackwards_withDurationLessAmr_doesNotAdjustSeek` previously used a custom 2,000ms `pollingWaitUntilCondition` helper to wait for playback to reach 500ms before seeking. On cold emulator starts, `MediaCodecList` initialization can exceed 2,000ms, causing the wait to time out while the uncancelled `postDelayed` polling loop continued running on the main thread and dereferenced a nulled player reference during `tearDown()`. Replace `pollingWaitUntilCondition` with `PlayerFence.passesContentPosition(500)` to match the rest of `CompositionPlayerSeekTest` and ensure listeners and callbacks are cleaned up when the test finishes. PiperOrigin-RevId: 994328318
Ensure `PreloadMediaSource.preload()` only calls `loadControl.onPrepared(PlayerId.PRELOAD)` on the first preload call, matching the single `loadControl.onReleased(PlayerId.PRELOAD)` call in `releasePreloadMediaSource()`. Issue: androidx#3445 PiperOrigin-RevId: 994333792
1. Why `maxOutputNits = 500.0` was problematic: Pure SDR output buffers (`[0.0, 1.0]` in BT.709) are display-referred relative to the OS brightness slider (`1.0` = `255 / 255` sRGB = 100% of current SDR UI white) and have no headroom above `1.0`. Previously, `bt2020DisplayLinearNitsToBt709DisplayLinear` normalized against `maxOutputNits = 500.0` (with an 80% electrical knee at `292.5` nits). Reserving `171-255 / 255` sRGB for `>203`-nit highlights baked a fixed `203.15 / 500.0 = 40.6%` linear attenuation into the entire `0-292.5` nit region (`171 / 255` sRGB at diffuse white). Because the viewer's eyes adapt to the surrounding `255 / 255` SDR UI white—which the display compositor scales uniformly with the brightness slider—this dimmed shadows, midtones, and diffuse white by ~60% in linear light and reduced perceived colorfulness across all screen brightness settings. 2. Why anchoring at `maxOutputNits = 203.1521` is better: Anchoring `maxOutputNits = 203.1521` at ITU-R BT.2408 diffuse white aligns `1.0` with the SDR reference white point expected by the display compositor, letting the OS brightness slider scale physical brightness uniformly with the UI. Placing the 80% electrical inflection point at `118.9` nits (`0.80^2.4 * 203.1521`): 3. Increase BT.2020 to BT.709 soft gamut compression knee from 0.80 to 0.95: Because max-RGB distance `d = (ach - C) / ach` is measured in linear light while sRGB encoding follows a `~1/2.4` power law, a linear knee at `t0 = 0.80` (`C = 0.20` -> `124 / 255` sRGB) preserved only `131 / 255` (`51.5%`) of the perceptual sRGB saturation span from white uncompressed, desaturating the remaining `124 / 255` (`48.5%`) span. Increasing the linear knee threshold to `t0 = 0.95` (`C = 0.05` -> `63 / 255` sRGB, `h = 0.05`, `a = 0.9223432`) preserves `192 / 255` (`75.2%`) of the perceptual sRGB saturation span at 1:1 without desaturation while softly rolling off out-of-gamut BT.2020 primaries into `[0.0, 1.0]`. PiperOrigin-RevId: 994338584
In `protobuf-gradle-plugin`, `task.plugins` configures external `protoc-gen-*` plugin binaries (such as `grpc`), whereas `task.builtins` configures code generators built directly into `protoc` (such as `java`). Putting `"java"` inside `task.plugins` causes the plugin to log: ``` protoc plugin 'java' not defined. Trying to use 'protoc-gen-java' from system path ``` Switch `task.plugins` to `task.builtins` so `protobuf-gradle-plugin` invokes the built-in `java` generator directly. See: - https://github.com/google/protobuf-gradle-plugin#configure-what-to-generate - https://developer.android.com/jetpack/androidx/releases/datastore#protobuf-serialization PiperOrigin-RevId: 994373928
When multiple HLS interstitials share the same AdGroup and one of them is an X-ASSET-LIST interstitial, resolving the asset list previously overwrote the entire AdGroup.contentResumeOffsetUs with the resume offset of the asset list interstitial alone. Instead, subtract the interstitial's prior unresolved contribution to the group's contentResumeOffsetUs (clamped at 0 to handle single-ad group resets) and add the newly resolved resume offset, preserving the contributions of any other interstitials in the same AdGroup. Also remove any stale pending snap-in resolution for the interstitial if its newly resolved duration no longer requires deferred snapping. PiperOrigin-RevId: 994391990
Test parameter fields annotated with `@BindFlag` or `@BindCanaryMode`
are frequently read only via reflection by `Media3FlagsRule`, which
triggers ErrorProne `UnusedVariable` warnings unless manually
suppressed on every field.
Meta-annotate `@BindFlag` and `@BindCanaryMode` with ErrorProne's
`@Keep` annotation so annotated fields are recognized as used without
requiring `@SuppressWarnings("unused")`.
PiperOrigin-RevId: 994396820
Add an overload for `PlayerStateObserver` and `Player.observeState` accepting a `stateUpdater: (Player, Player.Events) -> Unit` callback that triggers initially with empty `Player.Events` and forwards the `Player.Events` whenever one of the configured events occurs. Migrate `PresentationState` to use `player?.observeState` with this new overload, extracting `updateState(player, events)` to handle initial state sync and selectively update pixelWidthHeightRatio-adjusted video size, shutter visibility on first rendered frame, and track selections. PiperOrigin-RevId: 994436807
Route `maybeUpdateSourceInfo` through `ComponentListener` in `AdsMediaSource.onChildSourceInfoRefreshed` instead of storing a nullable `playerHandler` field on `AdsMediaSource`, avoiding a race condition where `releaseSourceInternal` clears the handler before the `mainHandler` callback runs. Issue: androidx#3442 PiperOrigin-RevId: 994459054
Unnamed variables (`_`) are a Java 22 feature and cannot be used in the Media3 Gradle build yet. Replace `boolean _` with `boolean unused` in `AudioTrackAudioOutputTest`. PiperOrigin-RevId: 994472016
Allow FragmentedMp4Muxer to emit duration-based fragments when video key frames are farther apart. Keep the default behavior unchanged and avoid non-key cuts when sample reordering is possible. Add focused tests for enabled and disabled behavior, reordered video, and empty samples.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
19 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.