ExoPlayer 源码阅读小记--HLS播放带缓存加载M38U文件过程

基于ExoPlayer 2.17.1源码分析，基本是一边看一边写的流水账，记录下防止以后忘了：

第一步createMediaSource创建HlsMediaSource对象时同时会实例化出HlsPlaylistTracker.Factory

mediaSource = new HlsMediaSource.Factory(getDataSourceFactoryCache(mAppContext, cacheEnable, preview, cacheDir, uerAgent)).setAllowChunklessPreparation(true).setPlaylistParserFactory(new HlsPlaylistParserFactory()).createMediaSource(mediaItem);

    public HlsMediaSource createMediaSource(MediaItem mediaItem) {checkNotNull(mediaItem.localConfiguration);HlsPlaylistParserFactory playlistParserFactory = this.playlistParserFactory;List<StreamKey> streamKeys = mediaItem.localConfiguration.streamKeys;if (!streamKeys.isEmpty()) {playlistParserFactory =new FilteringHlsPlaylistParserFactory(playlistParserFactory, streamKeys);}return new HlsMediaSource(mediaItem,hlsDataSourceFactory,extractorFactory,compositeSequenceableLoaderFactory,drmSessionManagerProvider.get(mediaItem),loadErrorHandlingPolicy,playlistTrackerFactory.createTracker(hlsDataSourceFactory, loadErrorHandlingPolicy, playlistParserFactory),//实例化TrackerFactoryelapsedRealTimeOffsetMs,allowChunklessPreparation,metadataType,useSessionKeys);}

第二步调用prepare(),在准备阶段将m3u8文件下载并解析

     player.prepare();

//ExoPlayerImpl
public void prepare() {....internalPlayer.prepare();......}

//ExoPlayerImplInternalprivate void prepareInternal() {....mediaSourceList.prepare(bandwidthMeter.getTransferListener());....}

//MediaSourceListpublic void prepare(@Nullable TransferListener mediaTransferListener) {....for (int i = 0; i < mediaSourceHolders.size(); i++) {MediaSourceHolder mediaSourceHolder = mediaSourceHolders.get(i);prepareChildSource(mediaSourceHolder);enabledMediaSourceHolders.add(mediaSourceHolder);}....}private void prepareChildSource(MediaSourceHolder holder) {.....mediaSource.prepareSource(caller, mediaTransferListener, playerId);}

//BaseMediaSource
public final void prepareSource(MediaSourceCaller caller,@Nullable TransferListener mediaTransferListener,PlayerId playerId) {....prepareSourceInternal(mediaTransferListener);....}

//HlsMediaSource
protected void prepareSourceInternal(@Nullable TransferListener mediaTransferListener) {....playlistTracker.start(localConfiguration.uri, eventDispatcher, /* primaryPlaylistListener= */ this);}

到这里使用HlsPlaylistTracker，这个类主要是管理播放列表（控制加载解析流程）,主列表叫primary playlist,另外该类还能获取当前播放的列表快照，该类入口就在这个start方法

//DefaultHlsPlaylistTracker
public void start(Uri initialPlaylistUri,EventDispatcher eventDispatcher,PrimaryPlaylistListener primaryPlaylistListener) {....ParsingLoadable<HlsPlaylist> multivariantPlaylistLoadable =new ParsingLoadable<>(dataSourceFactory.createDataSource(C.DATA_TYPE_MANIFEST),initialPlaylistUri,C.DATA_TYPE_MANIFEST,playlistParserFactory.createPlaylistParser());Assertions.checkState(initialPlaylistLoader == null);initialPlaylistLoader = new Loader("DefaultHlsPlaylistTracker:MultivariantPlaylist");//创建m3u8文件的网络加载解析线程long elapsedRealtime =initialPlaylistLoader.startLoading(multivariantPlaylistLoadable,this,loadErrorHandlingPolicy.getMinimumLoadableRetryCount(multivariantPlaylistLoadable.type));.....}

ParsingLoadable主要代码就这个方法，先通过DataSource获取需要加载的文件流，将文件流传入解析器解析，这里说下DataSourceInputStream本质是InputStream子类，但是里面包含一个datasSource和dataSpec，datasSource可以理解为数据源访问工具（如网络数据源访问Okhttp）获取到数据流后会将流保存在datasSource以供读取，dataSpec用来标记当前数据范围，里面还封装了读取数据的方式参数（如http请求url和header）

//ParsingLoadablepublic final void load() throws IOException {// We always load from the beginning, so reset bytesRead to 0.dataSource.resetBytesRead();DataSourceInputStream inputStream = new DataSourceInputStream(dataSource, dataSpec);try {inputStream.open();Uri dataSourceUri = Assertions.checkNotNull(dataSource.getUri());result = parser.parse(dataSourceUri, inputStream);} finally {Util.closeQuietly(inputStream);}}

当DataSourceInputStream调用open时，就会通过 dataSource.open(dataSpec)
当DataSourceInputStream，reader时就读取dataSource的数据流

//DataSourceInputStreamprivate void checkOpened() throws IOException {if (!opened) {dataSource.open(dataSpec);opened = true;}}public int read(byte[] buffer, int offset, int length) throws IOException {Assertions.checkState(!closed);checkOpened();int bytesRead = dataSource.read(buffer, offset, length);if (bytesRead == C.RESULT_END_OF_INPUT) {return -1;} else {totalBytesRead += bytesRead;return bytesRead;}}

好了我们继续open流程，StatsDataSource是一个封装DataSource的DataSource，里面包含一个DataSource 重定向的 uri 和响应标头等数据

//StatsDataSourcepublic long open(DataSpec dataSpec) throws IOException {// Reassign defaults in case dataSource.open throws an exception.lastOpenedUri = dataSpec.uri;lastResponseHeaders = Collections.emptyMap();long availableBytes = dataSource.open(dataSpec);lastOpenedUri = Assertions.checkNotNull(getUri());lastResponseHeaders = getResponseHeaders();return availableBytes;}

继续调用StatsDataSource封装的DataSource，继续套娃，套娃一般是为了分层，因为设置的是CacheDataSource这里会调用TeeDataSource 的open，Tees翻译就是三通的意思，这是个一进二出的三通，一个水龙头冷水进，一路进小厨宝缓存加热后出，一路直接水龙出，

//TeeDataSource
public long open(DataSpec dataSpec) throws IOException {bytesRemaining = upstream.open(dataSpec);//调用OkHttpDataSource open打开冷水进水阀门，准备一路上出水到水龙头，一路到小厨宝缓存if (bytesRemaining == 0) {return 0;}if (dataSpec.length == C.LENGTH_UNSET && bytesRemaining != C.LENGTH_UNSET) {// Reconstruct dataSpec in order to provide the resolved length to the sink.dataSpec = dataSpec.subrange(0, bytesRemaining);}dataSinkNeedsClosing = true;dataSink.open(dataSpec);//dataSink由CacheDataSource设置,调用CacheDataSink open打开厨宝的进水阀，准备下出水到小厨宝缓存return bytesRemaining;}

OkHttpDataSource 就忽略了，就是访问网络获取m3u8的文件流，这里看下CacheDataSink ,openNextOutputStream主要就是创建缓存的目录，创建缓存文件并获取输出流，这里并没有写入数据

//CacheDataSink
public void open(DataSpec dataSpec) throws CacheDataSinkException {....try {openNextOutputStream(dataSpec);} catch (IOException e) {throw new CacheDataSinkException(e);}}
private void openNextOutputStream(DataSpec dataSpec) throws IOException {....file =cache.startFile(castNonNull(dataSpec.key), dataSpec.position + dataSpecBytesWritten, length);FileOutputStream underlyingFileOutputStream = new FileOutputStream(file);if (bufferSize > 0) {if (bufferedOutputStream == null) {bufferedOutputStream =new ReusableBufferedOutputStream(underlyingFileOutputStream, bufferSize);} else {bufferedOutputStream.reset(underlyingFileOutputStream);}outputStream = bufferedOutputStream;} else {outputStream = underlyingFileOutputStream;}....}

那么在哪里什么时机写入缓存文件了呢,通过上面open的流程知道，open的流也没有立即读取而是缓存在DataSource里，当调用DataSource的read方法才会读取，所以理所应当写入缓存的时机就在读取的时候
，回到TeeDataSource read方法，具体的read时间点在下面会讲到，我们继续往下看

//TeeDataSourcepublic int read(byte[] buffer, int offset, int length) throws IOException {if (bytesRemaining == 0) {return C.RESULT_END_OF_INPUT;}int bytesRead = upstream.read(buffer, offset, length);if (bytesRead > 0) {// TODO: Consider continuing even if writes to the sink fail.这里还有个TODO作者想要缓存写入失败的时候也不要影响读取dataSink.write(buffer, offset, bytesRead);if (bytesRemaining != C.LENGTH_UNSET) {bytesRemaining -= bytesRead;}}return bytesRead;}

缓存相关先不展开说了，读取看完继续看解析，这里没啥好说的需要熟悉m3u8文件的格式，返回一个解析后的数据结构

//HlsPlaylistParser
public HlsPlaylist parse(Uri uri, InputStream inputStream) throws IOException {BufferedReader reader = new BufferedReader(new InputStreamReader(inputStream));Queue<String> extraLines = new ArrayDeque<>();String line;try {if (!checkPlaylistHeader(reader)) {//这里校验M3U8文件头，有些不规范的M3U8文件会包含BOM头校验会不通过，可以在创建MediaSource时通过setPlaylistParserFactory设置自定义HlsPlaylistParser修改支持BOM头throw ParserException.createForMalformedManifest(/* message= */ "Input does not start with the #EXTM3U header.", /* cause= */ null);}while ((line = reader.readLine()) != null) {line = line.trim();if (line.isEmpty()) {// Do nothing.} else if (line.startsWith(TAG_STREAM_INF)) {extraLines.add(line);return parseMultivariantPlaylist(new LineIterator(extraLines, reader), uri.toString());} else if (line.startsWith(TAG_TARGET_DURATION)|| line.startsWith(TAG_MEDIA_SEQUENCE)|| line.startsWith(TAG_MEDIA_DURATION)|| line.startsWith(TAG_KEY)|| line.startsWith(TAG_BYTERANGE)|| line.equals(TAG_DISCONTINUITY)|| line.equals(TAG_DISCONTINUITY_SEQUENCE)|| line.equals(TAG_ENDLIST)) {extraLines.add(line);return parseMediaPlaylist(multivariantPlaylist,previousMediaPlaylist,new LineIterator(extraLines, reader),uri.toString());} else {extraLines.add(line);}}} finally {Util.closeQuietly(reader);}throw ParserException.createForMalformedManifest("Failed to parse the playlist, could not identify any tags.", /* cause= */ null);}

解析完成又会回到DefaultHlsPlaylistTracker的onLoadCompleted，会获取刚才的额解析结果

//DefaultHlsPlaylistTracker
public void onLoadCompleted(ParsingLoadable<HlsPlaylist> loadable, long elapsedRealtimeMs, long loadDurationMs) {HlsPlaylist result = loadable.getResult();//获取解析结果.......MediaPlaylistBundle primaryBundle = playlistBundles.get(primaryMediaPlaylistUrl);if (isMediaPlaylist) {// We don't need to load the playlist again. We can use the same result.primaryBundle.processLoadedPlaylist((HlsMediaPlaylist) result, loadEventInfo);} else {primaryBundle.loadPlaylist();}....}private void processLoadedPlaylist(HlsMediaPlaylist loadedPlaylist, LoadEventInfo loadEventInfo) {@Nullable HlsMediaPlaylist oldPlaylist = playlistSnapshot;....if (playlistSnapshot != oldPlaylist) {playlistError = null;lastSnapshotChangeMs = currentTimeMs;onPlaylistUpdated(playlistUrl, playlistSnapshot);} ....}private void onPlaylistUpdated(Uri url, HlsMediaPlaylist newSnapshot) {if (url.equals(primaryMediaPlaylistUrl)) {....primaryMediaPlaylistSnapshot = newSnapshot;primaryPlaylistListener.onPrimaryPlaylistRefreshed(newSnapshot);//首次加载primary}
....}

playlist获取后就会交友MediaSource类处理，MediaSource 有两个主要职责：

为播放器提供Timeline(定义了媒体结构)，当在媒体结构发生变化时提供新的Timeline。MediaSource 通过调用MediaSource.MediaSourceCaller.onSourceInfoRefreshed来提供这些时间线。
为其时间线中的Period提供MediaPeriod实例。MediaPeriods 是通过调用createPeriod获得的，并为播放器提供了一种加载和读取媒体的方式。关于Timeline

//HlsMediaSource
public void onPrimaryPlaylistRefreshed(HlsMediaPlaylist mediaPlaylist) {....SinglePeriodTimeline timeline =playlistTracker.isLive()? createTimelineForLive(mediaPlaylist, presentationStartTimeMs, windowStartTimeMs, manifest): createTimelineForOnDemand(mediaPlaylist, presentationStartTimeMs, windowStartTimeMs, manifest);//Timeline在这里创建refreshSourceInfo(timeline);}

//BaseMediaSourceprotected final void refreshSourceInfo(Timeline timeline) {this.timeline = timeline;for (MediaSourceCaller caller : mediaSourceCallers) {caller.onSourceInfoRefreshed(/* source= */ this, timeline);}}

//CompositeMediaSource
protected final void prepareChildSource(@UnknownNull T id, MediaSource mediaSource) {,....MediaSourceCaller caller =(source, timeline) -> onChildSourceInfoRefreshed(id, source, timeline);//调用此处prepare时注册的caller....}

//MaskingMediaSource
protected void onChildSourceInfoRefreshed(Void id, MediaSource mediaSource, Timeline newTimeline) {@Nullable MediaPeriodId idForMaskingPeriodPreparation = null;....if (idForMaskingPeriodPreparation != null) {Assertions.checkNotNull(unpreparedMaskingMediaPeriod).createPeriod(idForMaskingPeriodPreparation);//创建HlsMediaPeriod}}
public void createPeriod(MediaPeriodId id) {....if (callback != null) {mediaPeriod.prepare(/* callback= */ this, preparePositionUs);}}

HlsMediaPeriod用来加载Timeline里的Period

//HlsMediaPeriodpublic void prepare(Callback callback, long positionUs) {this.callback = callback;playlistTracker.addListener(this);buildAndPrepareSampleStreamWrappers(positionUs);}private void buildAndPrepareSampleStreamWrappers(long positionUs) {....for (HlsSampleStreamWrapper sampleStreamWrapper : this.sampleStreamWrappers) {sampleStreamWrapper.continuePreparing();}....}

HlsSampleStreamWrapper包含一个HlsChunkSource用来获取HlsMediaChunk进行加载提供出SampleStream，
Chunk可以理解为数据块，这里对应一个ts文件，通过ChunkSource获取Chunk执行加载

//HlsSampleStreamWrapperpublic void continuePreparing() {if (!prepared) {continueLoading(lastSeekPositionUs);}}
public boolean continueLoading(long positionUs) {....chunkSource.getNextChunk(positionUs,loadPositionUs,chunkQueue,/* allowEndOfStream= */ prepared || !chunkQueue.isEmpty(),nextChunkHolder);
....@Nullable Chunk loadable = nextChunkHolder.chunk;....long elapsedRealtimeMs =loader.startLoading(//加载获取到的chunk，这里可以看到将回调函数设置在了HlsSampleStreamWrapper，后面加载完成后会回调它的onLoadCompleted方法loadable, this, loadErrorHandlingPolicy.getMinimumLoadableRetryCount(loadable.type));....return true;}

这里看下获取下一个chunk 的代码，其实就是从解析的playerlist里通过获取Segment创建出对应的chunk

//HlsChunkSource
public void getNextChunk(long playbackPositionUs,long loadPositionUs,List<HlsMediaChunk> queue,boolean allowEndOfStream,HlsChunkHolder out) {....// Select the track.MediaChunkIterator[] mediaChunkIterators = createMediaChunkIterators(previous, loadPositionUs);trackSelection.updateSelectedTrack(playbackPositionUs, bufferedDurationUs, timeToLiveEdgeUs, queue, mediaChunkIterators);....@NullableHlsMediaPlaylist playlist =playlistTracker.getPlaylistSnapshot(selectedPlaylistUrl, /* isForPlayback= */ true);//通过Tracker获取当前播放的列表
....Pair<Long, Integer> nextMediaSequenceAndPartIndex =getNextMediaSequenceAndPartIndex(previous, switchingTrack, playlist, startOfPlaylistInPeriodUs, loadPositionUs);//获取下一个Sequence也就是m3u8里的下一个片段索引long chunkMediaSequence = nextMediaSequenceAndPartIndex.first;int partIndex = nextMediaSequenceAndPartIndex.second;....@NullableSegmentBaseHolder segmentBaseHolder =getNextSegmentHolder(playlist, chunkMediaSequence, partIndex);//获取信息....//创建通过获取的Segment创建出对应的chunkout.chunk =HlsMediaChunk.createInstance(extractorFactory,mediaDataSource,playlistFormats[selectedTrackIndex],startOfPlaylistInPeriodUs,playlist,segmentBaseHolder,selectedPlaylistUrl,muxedCaptionFormats,trackSelection.getSelectionReason(),trackSelection.getSelectionData(),isTimestampMaster,timestampAdjusterProvider,previous,/* mediaSegmentKey= */ keyCache.get(mediaSegmentKeyUri),/* initSegmentKey= */ keyCache.get(initSegmentKeyUri),shouldSpliceIn,playerId);}

MediaChunk执行chunk加载，首先通过prepareExtraction创建出input数据流和HlsMediaChunkExtractor执行器，然后通过执行器执行相应的流，流的这块后续处理还是比较复杂的不是本文重点，主要目的就是ts文件的Demux，解析出Demux（对应代码里的Sample，相当于一帧）后的数据发送到一个DataQuene里供Renderer获取，Renderer最终调用系统的MediaCodec渲染到surface上

//HlsMediaChunk
public void load() throws IOException {....if (!loadCanceled) {if (!hasGapTag) {loadMedia();}loadCompleted = !loadCanceled;}
}private void loadMedia() throws IOException {feedDataToExtractor(dataSource, dataSpec, mediaSegmentEncrypted, /* initializeTimestampAdjuster= */ true);
}private void feedDataToExtractor(DataSource dataSource,DataSpec dataSpec,boolean dataIsEncrypted,boolean initializeTimestampAdjuster)throws IOException {....try {ExtractorInput input =prepareExtraction(dataSource, loadDataSpec, initializeTimestampAdjuster);if (skipLoadedBytes) {input.skipFully(nextLoadPosition);}try {while (!loadCanceled && extractor.read(input)) {}//通过执行器执行相应的流} catch (EOFException e) {....
}private DefaultExtractorInput prepareExtraction(DataSource dataSource, DataSpec dataSpec, boolean initializeTimestampAdjuster)throws IOException {long bytesToRead = dataSource.open(dataSpec);//这里又看到熟悉的代码类似ParsingLoadable加载m3u8文件一样，调用我们的三通TeeDataSource从网络获取ts文件，数据流会保存在datasource等待读取，读取时缓存文件....DefaultExtractorInput extractorInput =new DefaultExtractorInput(dataSource, dataSpec.position, bytesToRead);//数据流if (extractor == null) {....extractor =//执行器previousExtractor != null? previousExtractor.recreate(): extractorFactory.createExtractor(dataSpec.uri,trackFormat,muxedCaptionFormats,timestampAdjuster,dataSource.getResponseHeaders(),extractorInput,playerId);....return extractorInput;
}

到这里第一个ts文件已经加载，读取流，完成后会回到HlsSampleStreamWrapper这个类的onLoadCompleted方法，后续的ts文件加载会在播放器主loop里循环判断是否要继续加载，如果判断需要继续加载，最终调用continueLoading实现加载下一个ts文件。最具调用判断过程后续会讲到。

//HlsSampleStreamWrapper
public void onLoadCompleted(Chunk loadable, long elapsedRealtimeMs, long loadDurationMs) {loadingChunk = null;chunkSource.onChunkLoadCompleted(loadable);
....if (!prepared) {continueLoading(lastSeekPositionUs);} else {callback.onContinueLoadingRequested(this);}}

至此HLS的m3u8文件加载解析以及ts文件读取的大致流程就完成了

ExoPlayer 源码阅读小记--HLS播放带缓存加载M38U文件过程相关推荐

Soul网关源码阅读（九）插件配置加载初探
Soul网关源码阅读(九)插件配置加载初探简介今日来探索一下插件的初始化,及相关的配置的加载源码Debug 插件初始化首先来到我们非常熟悉的插件链调用的类: SoulWebHa ...
【GStreamer源码分析】playbin播放test.wav加载插件过程分析
playbin播放test.wav加载插件过程分析一.前言二.playbin 播放 .wav 音频插件加载一览三.测试代码 3.1 gst_init 3.2 gst_element_set_st ...
php源码自动识别文本中的链接,自动加载识别文件Auto.php
用于本应用的控制器自动加载类设置,用法如同\CodeIgniter\Config\AutoloadConfig 自动加载识别文件:dayrui/App/应用目录/Config/Auto.php 语法格 ...
sfm三维重建源码_OpenMVG源码阅读小记
"读一份好源码,就是和许多智慧的人谈话". 本文记录了笔者学习 openMVG 开源软件的一些初步经验和心得.如果你对计算机视觉和摄影测量有兴趣,需要用到相关技术,这篇文章正好就是 ...
java中talent-aio_talent-aio源码阅读小记（一）
近来在oschina上看到一个很火的java 即时通讯项目talent-aio,恰巧想了解一下这方面的东西,就阅读了一下项目的源码,这里对自己阅读源码后的一些心得体会做一下备忘,也希望能够对其他项目中 ...
Spring Ioc 源码分析(一)--Spring Ioc容器的加载
1.目标:熟练使用spring,并分析其源码,了解其中的思想.这篇主要介绍spring ioc 容器的加载 2.前提条件:会使用debug 3.源码分析方法:Intellj idea debug 模式 ...
linux dlopen 源码,采用dlopen、dlsym、dlclose加载动态链接库
采用dlopen.dlsym.dlclose加载动态链接库转载请标注,熬夜写的文章,挺辛苦 ... 环境系统: 16.04.1-Ubuntu 编译器: gnu 5.4.0 python: 2.7. ...
HTML卡片式布局源码,html5自适应卡片式设计动态加载整站源码_
html5自适应卡片式设计动态加载整站源码该模板是非常容易存活的,这样的程序很容易吸引访客点击,提升ip流量和pv是非常有利的,随意挂点联盟广告都能养活程序. 本套整站源码采使用现在非常流行的全屏自 ...
wemall app商城源码中基于JAVA的Android异步加载图片管理器代码
wemall doraemon是Android客户端程序,服务端采用wemall微信商城,不对原商城做任何修改,只需要在原商城目录下上传接口文件即可完成服务端的配置,客户端可随意定制修改.本文分享其中 ...

ExoPlayer 源码阅读小记--HLS播放带缓存加载M38U文件过程

ExoPlayer 源码阅读小记--HLS播放带缓存加载M38U文件过程相关推荐

最新文章

热门文章