一次“error domain=avfoundationerrordomain code=-11841”的调试

起因

最近在重构视频输出模块的时候，调试碰到avassetreader 调用开始方法总是返回no而失败，代码如下：

if ([reader startreading] == no) 
    {
        nslog(@"error reading from file at url: %@", self.url);
        return;
    }

reader的创建代码如下，主要用的是gpuimagemoviecompostion.

- (avassetreader*)createassetreader
{
    nserror *error = nil;
    avassetreader *assetreader = [avassetreader assetreaderwithasset:self.compositon error:&error];
    
    nsdictionary *outputsettings = @{(id)kcvpixelbufferpixelformattypekey: @(kcvpixelformattype_420ypcbcr8biplanarfullrange)};
    avassetreadervideocompositionoutput *readervideooutput = [avassetreadervideocompositionoutput assetreadervideocompositionoutputwithvideotracks:[_compositon trackswithmediatype:avmediatypevideo]
                                                                                                                                     videosettings:outputsettings];
    readervideooutput.videocomposition = self.videocomposition;
    readervideooutput.alwayscopiessampledata = no;
    if ([assetreader canaddoutput:readervideooutput]) {
        [assetreader addoutput:readervideooutput];
    }
    
    nsarray *audiotracks = [_compositon trackswithmediatype:avmediatypeaudio];
    bool shouldrecordaudiotrack = (([audiotracks count] > 0) && (self.audioencodingtarget != nil) );
    avassetreaderaudiomixoutput *readeraudiooutput = nil;
    
    if (shouldrecordaudiotrack)
    {
        [self.audioencodingtarget setshouldinvalidateaudiosamplewhendone:yes];
        nsdictionary *audioreadersetting = @{avformatidkey: @(kaudioformatlinearpcm),
                                             avlinearpcmisbigendiankey: @(no),
                                             avlinearpcmisfloatkey: @(no),
                                             avlinearpcmbitdepthkey: @(16)};
        readeraudiooutput = [avassetreaderaudiomixoutput assetreaderaudiomixoutputwithaudiotracks:audiotracks audiosettings:audioreadersetting];
        readeraudiooutput.audiomix = self.audiomix;
        readeraudiooutput.alwayscopiessampledata = no;
        [assetreader addoutput:readeraudiooutput];
    }
    
    return assetreader;
}

然后在该处断点，查看reader的status和error，显示error domain=avfoundationerrordomain code=-11841。查了一下文档发现如下：

averrorinvalidvideocomposition = -11841

you attempted to perform a video composition operation that is not supported.

应该就是readervideooutput.videocomposition = self.videocomposition; 这个videocompostion有问题了。avmutablevideocomposition这个类在视频编辑里面非常重要，包含对avcomposition中的各个视频track如何融合的信息，我做的是两个视频的融合小demo，就需要用到它，设置这个类稍微有点复杂，比较容易出错，特别是设置avmutablevideocompositioninstruction这个类的timerange属性，我的这次错误就是因为这个。

开始调试

马上调用avmutablevideocomposition的如下方法：

bool isvalid = [self.videocomposition isvalidforasset:self.compositon timerange:cmtimerangemake(kcmtimezero, self.compositon.duration) validationdelegate:self];

这个时候isvalid为no，确定就是这个videocompostion问题了，添加了avvideocompositionvalidationhandling协议四个方法打印一下，这个协议太贴心了，估计知道这个地方出错概率比较高，所以特地弄的吧。代码如下：

- (bool)videocomposition:(avvideocomposition *)videocomposition shouldcontinuevalidatingafterfindinginvalidvalueforkey:(nsstring *)key 
{
    nslog(@"%s===%@",__func__,key);
    return yes;
}

- (bool)videocomposition:(avvideocomposition *)videocomposition shouldcontinuevalidatingafterfindingemptytimerange:(cmtimerange)timerange 
{
    nslog(@"%s===%@",__func__,cfbridgingrelease(cmtimerangecopydescription(kcfallocatordefault, timerange)));
    return yes;
}

- (bool)videocomposition:(avvideocomposition *)videocomposition shouldcontinuevalidatingafterfindinginvalidtimerangeininstruction:(id<avvideocompositioninstruction>)videocompositioninstruction 
{
    nslog(@"%s===%@",__func__,videocompositioninstruction);
    return yes;
}

- (bool)videocomposition:(avvideocomposition *)videocomposition shouldcontinuevalidatingafterfindinginvalidtrackidininstruction:(id<avvideocompositioninstruction>)videocompositioninstruction layerinstruction:(avvideocompositionlayerinstruction *)layerinstruction asset:(avasset *)asset 
{
    nslog(@"%s===%@===%@",__func__,layerinstruction,asset);
    return yes;
}

重新运行一下发现第三个方法打印输出，就是instruction的timerange问题了，又重新看了一下两个instruction的设置问题,感觉又没什么问题，代码如下：

- (void)buildcompositionwithassets:(nsarray *)assetsarray
{
    for (int i = 0; i < assetsarray.count; i++) {
        avurlasset *asset = assetsarray[i];
        nsarray *videotracks = [asset trackswithmediatype:avmediatypevideo];
        nsarray *audiotracks = [asset trackswithmediatype:avmediatypeaudio];
        
        avassettrack *videotrack = videotracks[0];
        avassettrack *audiotrack = audiotracks[0];
        nserror *error = nil;
        avmutablecompositiontrack *videot = [self.avcompostions.mutablecomps addmutabletrackwithmediatype:avmediatypevideo preferredtrackid:kcmpersistenttrackid_invalid];
        avmutablecompositiontrack *audiot = [self.avcompostions.mutablecomps addmutabletrackwithmediatype:avmediatypeaudio preferredtrackid:kcmpersistenttrackid_invalid];
        [videot inserttimerange:videotrack.timerange oftrack:videotrack attime:self.offsettime error:&error];
        [audiot inserttimerange:audiotrack.timerange oftrack:audiotrack attime:self.offsettime error:nil];
        nsassert(!error, @"insert error = %@",error);
        avmutablevideocompositioninstruction *instruction = [avmutablevideocompositioninstruction videocompositioninstruction];
        avmutablevideocompositionlayerinstruction *layerinstruction = [avmutablevideocompositionlayerinstruction videocompositionlayerinstructionwithassettrack:videot];
        instruction.layerinstructions = @[layerinstruction];
        instruction.timerange = cmtimerangemake(self.offsettime, asset.duration);
        [self.instrucionarray addobject:instruction];
        self.offsettime = cmtimeadd(self.offsettime,asset.duration);
    }
}

这个数组里面会有两个加载好的avasset对象，加载avasset的代码如下：

- (void)loadassetfrompath:(nsarray *)paths
{
    nsmutablearray *assetsarray = [nsmutablearray arraywithcapacity:paths.count];
    dispatch_group_t dispatchgroup = dispatch_group_create();
    for (int i = 0; i < paths.count; i++) {
        nsstring *path = paths[i];
        //first find from cache
        avasset *asset = [self.assetscache objectforkey:path];
        if (!asset) {
            nsdictionary *inputoptions = [nsdictionary dictionarywithobject:[nsnumber numberwithbool:yes] forkey:avurlassetpreferprecisedurationandtimingkey];
            asset = [avurlasset urlassetwithurl:[nsurl fileurlwithpath:path] options:inputoptions];
            // cache asset
            nsassert(asset != nil, @"can't create asset from path", path);
            nsarray *loadkeys = @[@"tracks", @"duration", @"composable"];
            [self loadasset:asset withkeys:loadkeys usingdispatchgroup:dispatchgroup];
            [self.assetscache setobject:asset forkey:path];
        }else {
        }
        [assetsarray addobject:asset];
    }
    dispatch_group_notify(dispatchgroup, dispatch_get_main_queue(), ^{
        !self.assetloadblock?:self.assetloadblock(assetsarray);
        
    });
}


- (void)loadasset:(avasset *)asset withkeys:(nsarray *)assetkeystoload usingdispatchgroup:(dispatch_group_t)dispatchgroup
{
    dispatch_group_enter(dispatchgroup);
    [asset loadvaluesasynchronouslyforkeys:assetkeystoload completionhandler:^(){
        for (nsstring *key in assetkeystoload) {
            nserror *error;
            
            if ([asset statusofvalueforkey:key error:&error] == avkeyvaluestatusfailed) {
                nslog(@"key value loading failed for key:%@ with error: %@", key, error);
                goto bail;
            }
        }
        if (![asset iscomposable]) {
            nslog(@"asset is not composable");
            goto bail;
        }
    bail:
        dispatch_group_leave(dispatchgroup);
    }];
}

按照正常来说，应该不会有什么问题的，但是问题还是来了。打印出创建的两个avmutablevideocompositioninstruction的信息，发现他们的timerange确实有重合的地方。avmutablevideocompositioninstruction的timerange必须对应avmutablecompositiontrack里面的一段段插入的track的timerange。开发文档是这么说的：

to report a video composition instruction with a timerange that's invalid, that overlaps with the timerange of a prior instruction, or that contains times earlier than the timerange of a prior instruction

我在插入的时候使用的是videotrack.timerange，而设置instruction.timerange = cmtimerangemake(self.offsettime, asset.duration);使用的是asset.duration。这个两个竟然是不一样的。asset.duration是{59885/1000 = 59.885}，videotrack.timerange是{{0/1000 = 0.000}, {59867/1000 = 59.867}}
这两个时长有细微差别，到底哪个是比较准确一点的呢？

视频文件时长的计算

笔者在demo中使用的是mp4格式的文件，mp4文件是若干个不同的类型box组成的，box可以理解为装有数据的容器。其中有一种moov类型的box里面装有视频播放的元数据（metadata），这里面有视频的时长信息：timescale和duration。 duration / timescale = 可播放时长（s）。从mvhd中读到timescal为0x03e8，duration为0xe9ed，也就是59885 / 1000，为59.885。再看看tkhd box里面的内容，这里面是包含单一track的信息。如下图：

从上图可以看出videotrack和audiotrack两个的时长是不一样的，videotrack为0xe9db，也就是59.867，audiotrack为0xe9ed，就是59.885。所以我们在计算timerange的时候最好统一使用相对精确一点的videotrack，而不要avasset的duration，尽量避免时间上的误差，视频精细化的编辑，对这些误差敏感。