原文:http://galoisplusplus.coding.me/blog/2014/07/30/memory-management-in-cocos2d-x-v3/
cocos2d-x移植自Objective C的cocos2d,其内存管理其实也来自于OC。因而对于写过OC程序的朋友来讲,cocos2d-x的内存管理应该是一目了然的,但对于本渣这枚没接触过OC的C++码农来说,或许直接看cocos2d-x源代码才是最直接快捷的方式。
Node类
我们首先来看Node类的代码,Node是cocos2d-x中极重要的基类,许多常用的Scene、Layer、MenuItem等都继承自Node。
Node的创建是通过以下的接口,该函数返回一个Node的静态对象指针:
下面让我们来看这个函数的实现。该函数采用二段式创建的方式——首先用new operator在heap中开辟空间并进行简单的初始化,假如new返回一个合法地址(cocos2d-x没有采用c++的异常处理机制),则接着init函数用于实际初始化Node的成员。只有在这二者都成功后,才把创建的指针设为autorelease
(关于autorelease
后面会继续解释)并返回。
Node::create() { ret = new Node(); if (ret && ret->init()) { autorelease(); } else { CC_SAFE_DELETE(ret); } return ret; }
对于创建失败的情况,cocos2d-x使用了下面的宏保证该指针被delete且被设为nullptr:
#define CC_SAFE_DELETE(p) do { delete (p); (p) = nullptr; } while(0)
这个二段式的create函数在cocos2d-x中非常常用,因而cocos2d-x用了以下一个叫CREATE_FUNC
来表示这个函数以便给继承Node的子类使用:
* define a create function for a specific type,such as Layer * @param \__TYPE__ class type to add create(),117)!important"> */ #define CREATE_FUNC(__TYPE__) \ static __TYPE__* create() \ { \ __TYPE__ *pRet = new __TYPE__(); \ if (pRet && pRet->init()) \ { \ pRet->autorelease(); \ return pRet; \ } \ else \ delete pRet; \ pRet = NULL; \ return NULL; \ }
这样,继承Node的子类(例如ExampleLayer
)只需要在类声明(class declaration)中加入CREATE_FUNC(类名)
(例如CREATE_FUNC(ExampleLayer)
),再override下init函数即可。
Ref类
在cocos2d-x中,Node类的父类是Ref类,之前我们所看到的autorelease
方法实际上就来自于这个父类。
下面我们先来看Ref类的声明,这里为了突出重点,我们忽略script binding的情况:
class CC_DLL Ref { public: * Retains the ownership. * * This increases the Ref's reference count. * @see release,autorelease * @js NA */ * 拿到所有权 * 这会增加引用计数 */ @H_403_596@void retain(); * Releases the ownership immediately. * This decrements the Ref's reference count. * If the reference count reaches 0 after the descrement,this Ref is * destructed. * @see retain,117)!important"> * 立即释放所有权 * 这会减少引用计数 * 如果更新后的引用计数为0,该Ref对象会被销毁 release(); * Releases the ownership sometime soon automatically. * This descrements the Ref's reference count at the end of current * autorelease pool block. * @returns The Ref itself. * @see AutoreleasePool,retain,release * @lua NA * 自动释放所有权 */ Ref* autorelease(); * Returns the Ref's current reference count. * @returns The Ref's reference count. * 返回该Ref对象的引用计数 unsigned @H_403_596@int getReferenceCount() const; protected: * Constructor * The Ref's reference count is 1 after construction. * 构造函数 * 初始引用计数为1 Ref(); */ virtual ~ * 采用引用计数(reference counting) * _referenceCount就是计数值 */ // count of references _referenceCount; friend AutoreleasePool; // Memory leak diagnostic data (only included when CC_USE_MEM_LEAK_DETECTION is defined and its value isn't zero) // 以下函数用于开启内存泄露检测时打印出泄露信息 #if CC_USE_MEM_LEAK_DETECTION public: static printLeaks(); #endif };
从上面的代码,我们可以初步了解到:Ref采用引用计数(reference counting)的方法来管理某个指针所指向的某个对象,初始创建时计数是1,当计数变为0时该对象被析构;retain
方法会增加计数并拿到所有权,而与之对应的,release
方法会减少计数;autorelease
是把所有权交给友类(friend class)AutoreleasePool
,让它来决定何时减少计数,这个类我们后面会继续谈到。
下面我们来看Ref类的实现(definition):
trackRef(ref); untrackRef(ref); #endif // 在初始化列表中将计数设为1 Ref::Ref() : _referenceCount(1) // when the Ref is created,the reference count of it is 1 { // 假如开启内存泄露检测,则追踪该对象指针,将该对象指针放入一个列表(list)中 // 后面的代码我们很快就会看到这个list #if CC_USE_MEM_LEAK_DETECTION trackRef(this); #endif } Ref::~Ref() { // 假如开启内存泄露检测且引用计数非0,则在追踪列表中找到该对象指针并删除 #if CC_USE_MEM_LEAK_DETECTION _referenceCount != 0) untrackRef(#endif } // retain只是单纯将计数递增 retain() { // CCASSERT是cocos2d-x对C++的assert所封装的宏 CCASSERT(_referenceCount > 0, "reference count should greater than 0"); ++_referenceCount; } release() { // 首先计数递减 @H_301_1128@"reference count should greater than 0"); --_referenceCount; // 计数为0,应当析构对象 _referenceCount == 0) { #if defined(COCOS2D_DEBUG) && (COCOS2D_DEBUG > 0) // 得到一个PoolManager单例的对象 // PoolManager类后面会解释 auto poolManager = PoolManager::getInstance(); // 后面会详细解释这段代码 if (!poolManager->getCurrentPool()->isClearing() && isObjectInPools(this)) { // 以下的注释很重要,很快会解释到 // Trigger an assert if the reference count is 0 but the Ref is still in autorelease pool. // This happens when 'autorelease/release' were not used in pairs with 'new/retain'. // // Wrong usage (1): // auto obj = Node::create(); // Ref = 1,but it's an autorelease Ref which means it was in the autorelease pool. // obj->autorelease(); // Wrong: If you wish to invoke autorelease several times,you should retain `obj` first. // Wrong usage (2): // auto obj = Node::create(); // obj->release(); // Wrong: obj is an autorelease Ref,it will be released when clearing current pool. // Correct usage (1): // |- new Node(); // `new` is the pair of the `autorelease` of next line // |- autorelease(); // The pair of `new Node`. // obj->retain(); // obj->autorelease(); // This `autorelease` is the pair of `retain` of prevIoUs line. // Correct usage (2): // obj->release(); // This `release` is the pair of `retain` of prevIoUs line. CCASSERT(false,152)!important">"The reference shouldn't be 0 because it is still in autorelease pool."); } // 假如开启内存泄露检测,则在追踪列表中找到该对象指针并删除 #endif // 调用析构函数并释放空间 delete this; } } // 把该对象指针交给友类AutoreleasePool(具体来说,是PoolManager单例对象所得到的当前的AutoreleasePool)来管理 autorelease() { getInstance()->addObject(this); return this; } const { _referenceCount; } #if CC_USE_MEM_LEAK_DETECTION // 这里便是存放所追踪的对象指针的列表 std::list<Ref*> __refAllocationList; printLeaks() { // Dump Ref object memory leaks __refAllocationList.empty()) { log("[memory] All Ref objects successfully cleaned up (no leaks detected).\n"); } "[memory] WARNING: %d Ref objects still active in memory.", (@H_403_596@int)size()); // C++的range-for语法 // 打印出每个泄露内存的对象指针的类型和引用计数 for (const auto& ref : __refAllocationList) { CC_ASSERT(ref); const @H_403_596@char* type = typeid(*ref).name(); @H_301_1128@"[memory] LEAK: Ref object '%s' still active with reference count %d.(type ? type : ""), ref->getReferenceCount()); } } } // 将对象指针放入列表中 ref) { ref,152)!important">"Invalid parameter,ref should not be null!"); // Create memory allocation record. push_back(ref); } // 在列表中找到该对象指针并删除 ref) { iter = find(begin(),210)!important">end(),210)!important">iter == end()) { @H_301_1128@"[memory] CORRUPTION: Attempting to free (%s) with invalid ref tracking record.name()); return; } erase(iter); } #endif // #if CC_USE_MEM_LEAK_DETECTION
-
当Ref的计数变为0时,它一定不能在AutoreleasePool中。
-
Ref的计数为0且同时在AutoreleasePool中的错误是由new/retain和autorelease/release没有对应引起的(有木有想起C++中new和delete没对应所引起的内存泄露?):
- autorelease缺乏对应的retain。 例如:
obj = create(); // 注意create函数会调用autorelease方法,因此obj已经没有该指针的所有权了 obj->autorelease(); // obj没有所有权,因此无法再把所有权转交给AutoreleasePool,若要调用autorelease方法需要先调用retain拿到所有权
- release缺乏对应的retain。
例如:
// 前面我们分析过create函数,它会先用new operator得到对象,再调用autorelease方法 // 这里new和autorelease对应 create(); |- Node(); |- autorelease(); // 这里retain和autorelease对应,autorelease一个已经被autorelease过的对象(例如通过create函数构造的对象)必须先retain retain(); autorelease();
又如:
create(); // 这里retain和release对应,release一个已经被autorelease过的对象(例如通过create函数构造的对象)必须先retain release();
AutoreleasePool类
现在我们来看Ref类的友类AutoreleasePool。 首先来看类声明:
AutoreleasePool { * @warn Don't create an auto release pool in heap,create it in stack. * 警告:不要在heap上构造AutoreleasePool对象,要在stack上构造 AutoreleasePool(); * Create an autorelease pool with specific name. This name is useful for debugging. AutoreleasePool(const string &name); */ ~ * Add a given object to this pool. * The same object may be added several times to the same pool; When the * pool is destructed,the object's Ref::release() method will be called * for each time it was added. * @param object The object to add to the pool. * 把指定的对象指针放到AutoreleasePool对象中 * 注意: * 同一对象的指针可能会被多次加入到同一AutoreleasePool对象中; * 当该AutoreleasePool对象被析构时,该对象指针被加入多少次,就得调用多少次该对象的release()函数 * 这是因为AutoreleasePool用vector而非set来存放所管理的对象指针,因此不会去重 addObject(Ref *object); * Clear the autorelease pool. * Ref::release() will be called for each time the managed object is * added to the pool. * 清空AutoreleasePool * 每个被管理的对象指针被加入多少次,就会调用多少次release()函数 clear(); #if defined(COCOS2D_DEBUG) && (COCOS2D_DEBUG > 0) * Whether the pool is doing `clear` operation. bool isClearing() const { _isClearing; }; #endif * Checks whether the pool contains the specified object. * 检查AutoreleasePool对象是否管理某个对象指针 contains(object) const; * Dump the objects that are put into autorelease pool. It is used for debugging. * The result will look like: * Object pointer address object id reference count dump(); private: * The underlying array of object managed by the pool. * Although Array retains the object once when an object is added,proper * Ref::release() is called outside the array to make sure that the pool * does not affect the managed object's reference count. So an object can * be destructed properly by calling Ref::release() even if the object * is in the pool. * AutoreleasePool对象将它所管理的对象指针放到下面的vector中 * 尽管每次有对象指针加到该vector中时,该vector实际上retain拿到了所有权, * 但是Ref::release()会被调用来保证AutoreleasePool不会改变它所管理的对象指针 * 的引用计数。 * 所以,当某个对象指针被放到AutoreleasePool类中管理时,仍然可以通过调用 * Ref::release()函数来析构它 vector<_managedObjectArray; string _name; * The flag for checking whether the pool is doing `clear` operation. _isClearing; 从类声明中能解读出的最重要的信息是AutoreleasePool类用STL vector来存放它所管理的Ref所指向的对象。要搞清楚原理还需要继续看它的实现:77
AutoreleasePool::AutoreleasePool() : _name("") #if defined(COCOS2D_DEBUG) && (COCOS2D_DEBUG > 0) ,210)!important">_isClearing(false) #endif { _managedObjectArray.reserve(150); // 每个新创建的AutoreleasePool对象都交由PoolManager单例对象统一管理 push(this); } name) : _name(name) AutoreleasePool::~AutoreleasePool() { CCLOGINFO("deallocing AutoreleasePool: %p",22)!important">this); // 清空该AutoreleasePool clear(); // 要析构的AutoreleasePool对象不再由PoolManager管理 pop(); } // 只是单纯调用vector::push_back加入所管理的对象 object) { object); } // clear函数就是AutoreleasePool调用release来管理对象的引用计数的地方 clear() { #if defined(COCOS2D_DEBUG) && (COCOS2D_DEBUG > 0) _isClearing = true; // 调用每个在AutoreleasePool的对象指针的release方法 auto &obj : _managedObjectArray) { release(); } // 清空存放管理对象的vector clear(); false; // 线性搜索所管理的对象指针的vector,查看所指定的Ref指针是否存在 _managedObjectArray) { obj == object) return true; } false; } dump() { CCLOG("autorelease pool: %s,number of managed object %d_name.c_str(),22)!important">static_cast<@H_403_596@int>(size())); @H_301_1128@"%20s%20s%20s",152)!important">"Object pointer",152)!important">"Object id",152)!important">"reference count"); CC_UNUSED_PARAM(obj); @H_301_1128@"%20p%20uobj,210)!important">getReferenceCount()); } }
PoolManager类
下面我们来看PoolManager类,在cocos2d-x中,这个类是典型的单例(singleton)工厂类——及有且只有一个PoolManager对象,该PoolManger有一个存放AutoreleasePool对象指针的stack,该stack是由STL::vector实现的。需要注意的是,cocos2d-x的单例类都不是线程安全的,跟内存管理紧密相关的PoolManager类也不例外,因此在多线程中使用cocos2d-x的接口需要特别注意内存管理的问题。
我们先来看类声明:
PoolManager { CC_DEPRECATED_ATTRIBUTE PoolManager* sharedPoolManager() { getInstance(); } getInstance(); purgePoolManager() { destroyInstance(); } destroyInstance(); * Get current auto release pool,there is at least one auto release pool that created by engine. * You can create your own auto release pool at demand,which will be put into auto releae pool stack. AutoreleasePool *getCurrentPool() const; isObjectInPools(obj) AutoreleasePool; private: // singleton类把构造函数和析构函数设为private,避免被调用 PoolManager(); ~PoolManager(); push(pool); pop(); s_singleInstance; // 同样用vector来存放所管理AutoreleasePool对象指针的列表 AutoreleasePool*> _releasePoolStack; };
再来看类实现:
s_singleInstance = nullptr; getInstance() { s_singleInstance == nullptr) { s_singleInstance = PoolManager(); // Add the first auto release pool AutoreleasePool("cocos2d autorelease pool"); } s_singleInstance; } destroyInstance() { delete s_singleInstance; nullptr; } PoolManager() { _releasePoolStack.@H_301_1128@10); } PoolManager::~@H_301_1128@"deallocing PoolManager: %p",22)!important">this); // 逐个析构所管理的AutoreleasePool对象 while (!AutoreleasePool* pool = back(); pool; } } // 加入AutoreleasePool对象指针时用的是stl::vector的push_back函数, // 于是调用back函数就可以得到最新被加入的AutoreleasePool对象指针 back(); } // 线性搜索每个被管理的AutoreleasePool, // 每个AutoreleasePool对象再用contains函数线性搜索一遍 pool : _releasePoolStack) { pool->obj)) pool) { pool); } pop() { CC_ASSERT(!empty()); pop_back(); }
最后的疑问
想必各位用惯了c++的看官在看完了以上的代码之后,最有疑问的还是神秘的Ref::autorelease
函数。我们从AutoreleasePool的源代码看到,事实上被autorelease的对象最后还是通过release函数来减少其引用计数的,只不过release函数不是由使用者来调用,而是AutoreleasePool来调用,调用的地方在AutoreleasePool::clear()
函数。那么AutoreleasePool如何个「auto」自动管理内存法儿?AutoreleasePool::clear()
会在哪个地方被调用?
谜底隐藏在cocos/base/CCDirector.cpp
中:
DisplayLinkDirector::mainLoop() { _purgeDirectorInNextLoop) { _purgeDirectorInNextLoop = false; purgeDirector(); } else if (! _invalid) { drawScene(); // release the objects clear(); } }
这里就不纠缠Director类的实现细节了,上面的代码揭示的事实是:在图像渲染的主循环中,如果当前的图形对象是在当前帧,则调用显示函数,并调用AutoreleasePool::clear()
减少这些对象的引用计数。mainLoop是每一帧都会自动调用的,所以下一帧时这些对象都被当前的AutoreleasePool对象release了一次。这也是AutoreleasePool「自动」的来由。