【GPUImage】同步队列死锁问题解决-借助dispatch_queue_set_specific

不要使用 dispatch_get_current_queue()

当然这个方法在iOS6.0就废弃了，这里主要是重温一下同步、异步以及死锁。

- (NSString *)name
{
    __block NSString *name;
    dispatch_sync(_syncQueue, ^{
        name = @"不服不行";
    });
    return name;
}

上面的这种写法可能会出现死锁，假如调用获取方法的队列恰好是同步操作所在的队列。那么dispatch_sync就会一直不返回，需要等待块执行完毕。
这里就想到使用dispatch_get_current_queue来解决问题。

- (NSString *)name
{
    __block NSString *name;
    void (^block)() = ^{
        name = @"iOS";
    };
    
    if (dispatch_get_current_queue() == _syncQueue) {
        block();
    }else {
        dispatch_sync(_syncQueue, block);
    }
    
    return name;
}

但是 dispatch_get_current_queue 仍然可能造成死锁

  dispatch_sync(_syncQueueA, ^{
      dispatch_sync(_syncQueueB, ^{
          void (^block)() = ^{
            
          };
          if (dispatch_get_current_queue() == _syncQueueA) {
              block();
          }else {
              dispatch_sync(_syncQueueA, block);
          }
      });
   });

dispatch_get_current_queue 由于返回的是当前队列，上面返回的是_syncQueueB。因此，针对_syncQueueA的同步派发依然会执行，所以还是会造成死锁。
其实要解决这个问题，可以使用队列特有数据 dispatch_queue_set_specific。只要去看源码就会发现 GPUImage(https://github.com/BradLarson/GPUImage) 等多数框架都是这么解决的。

GPUImage中如何做到 统一在某个队列执行 并防止死锁的

@implementation GPUImageContext

static void *openGLESContextQueueKey;

- (id)init;
{
    if (!(self = [super init]))
    {
		return nil;
    }

	openGLESContextQueueKey = &openGLESContextQueueKey;
    _contextQueue = dispatch_queue_create("com.sunsetlakesoftware.GPUImage.openGLESContextQueue", NULL);
    
#if OS_OBJECT_USE_OBJC
	dispatch_queue_set_specific(_contextQueue, openGLESContextQueueKey, (__bridge void *)self, NULL);
#endif
    shaderProgramCache = [[NSMutableDictionary alloc] init];
    shaderProgramUsageHistory = [[NSMutableArray alloc] init];
    
    return self;
}

+ (void *)contextKey {
	return openGLESContextQueueKey;
}


// GPUImage/Output/GPUImageOutput.m

void runSynchronouslyOnVideoProcessingQueue(void (^block)(void))
{
    dispatch_queue_t videoProcessingQueue = [GPUImageContext sharedContextQueue];
#if !OS_OBJECT_USE_OBJC
#pragma clang diagnostic push
#pragma clang diagnostic ignored "-Wdeprecated-declarations"
    if (dispatch_get_current_queue() == videoProcessingQueue)
#pragma clang diagnostic pop
#else
	if (dispatch_get_specific([GPUImageContext contextKey]))
#endif
	{
		block();
	}else
	{
        if(videoProcessingQueue)
            dispatch_sync(videoProcessingQueue, block);
        else
        {
//            NSLog(@"error: videoProcessingQueue should not be null");
        }
	}
}

void runAsynchronouslyOnVideoProcessingQueue(void (^block)(void))
{
    dispatch_queue_t videoProcessingQueue = [GPUImageContext sharedContextQueue];
    
#if !OS_OBJECT_USE_OBJC
#pragma clang diagnostic push
#pragma clang diagnostic ignored "-Wdeprecated-declarations"
    if (dispatch_get_current_queue() == videoProcessingQueue)
#pragma clang diagnostic pop
#else
    if (dispatch_get_specific([GPUImageContext contextKey]))
#endif
    {
        block();
    }else
    {
        dispatch_async(videoProcessingQueue, block);
    }
}

参考：

https://www.jianshu.com/p/a9066dde1f2b/

https://juejin.cn/post/7133204939475517477#heading-2