AutoReleasePool 工作机制

AutoReleasePool 工作机制

背景：

相比很多人都有这个困惑，@autoreleasepool 到底做了什么，今天我们就从源码的基础上去分析，彻底揭开它神秘面纱

源码路径：

https://opensource.apple.com/tarballs/objc4/

得益于objc 开源，我们可以一窥其本质，在接下来分析中我会尽可能引入源码

@autoreleasepool 转成c++到底是什么样的

Objective-C   @autoreleasepool     {         NSObject *pObjecdt = nil;         pObjecdt = [[Test alloc] init];         [pObjecdt autorelease];         NSLog(@"refcount = %d", CFGetRetainCount(pObjecdt));     }

OC 转 C++

 使用命令

xcrun -sdk iphoneos clang -arch arm64 -rewrite-objc main.m -o main-arm64.cpp

就可以把对对应m 转成c++

Objective-C  {          __AtAutoreleasePool __autoreleasepool;         NSObject *pObjecdt = __null;         pObjecdt = ((Test *(*)(id, SEL))(void *)objc_msgSend)((id)((Test *(*)(id, SEL))(void *)objc_msgSend)((id)objc_getClass("Test"), sel_registerName("alloc")), sel_registerName("init"));         ((NSObject *(*)(id, SEL))(void *)objc_msgSend)((id)pObjecdt, sel_registerName("autorelease"));         NSLog((NSString *)&__NSConstantStringImpl__var_folders_x4_q4k0pbd17n90b9x6_v_h_rb40000gn_T_main_092efe_mi_0, CFGetRetainCount(pObjecdt));     }

也就是说@autoreleasepool 转换成了__AtAutoreleasePool __autoreleasepool;

下边就介绍__AtAutoreleasePool的定义

__AtAutoreleasePool

C++ struct __AtAutoreleasePool {   __AtAutoreleasePool() {atautoreleasepoolobj = objc_autoreleasePoolPush();}   ~__AtAutoreleasePool() {objc_autoreleasePoolPop(atautoreleasepoolobj);}   void * atautoreleasepoolobj; };

通过上边可以看的出来 这是很简单的结构图，遵循的RAII的原则，也就是说构造的时候生成atautoreleasepoolobj 指针，而 西沟的时候把atautoreleasepoolobj 传入到objc_autoreleasePoolPop中去，所以要搞清楚AutoReleasePool 就要搞清楚objc_autoreleasePoolPush 与 objc_autoreleasePoolPop 是干啥的。

objc_autoreleasePoolPush

C++ struct AutoreleasePoolPageData {         magic_t const magic;         __unsafe_unretained id *next;         pthread_t const thread;         AutoreleasePoolPage * const parent;         AutoreleasePoolPage *child;         uint32_t const depth;         uint32_t hiwat; }; class AutoreleasePoolPage : private AutoreleasePoolPageData { } void * objc_autoreleasePoolPush(void) {     return AutoreleasePoolPage::push(); } #   define POOL_BOUNDARY nil     static inline void *push()     {         id *dest;         dest = autoreleaseFast(POOL_BOUNDARY);         return dest;     }         //obj = POOL_BOUNDARY        static inline id *autoreleaseFast(id obj)         {             AutoreleasePoolPage *page = hotPage();             if (page && !page->full()) {                 return page->add(obj);             } else if (page) {                 return autoreleaseFullPage(obj, page);             } else {                 return autoreleaseNoPage(obj);             }         }                         static inline AutoreleasePoolPage *hotPage()             {                 AutoreleasePoolPage *result = (AutoreleasePoolPage *)                     tls_get_direct(key);                 if ((id *)result == EMPTY_POOL_PLACEHOLDER) return nil;                 if (result) result->fastcheck();                 return result;             }             class AutoreleasePoolPage             {                 //obj = POOL_BOUNDARY                 id *add(id obj)                 {                                 id *ret;                                               ret = next;  // faster than `return next-1` because of aliasing                     *next++ = obj;                                 return ret;                 }             }

通过上边调用关系可以看的出来，AutoreleasePoolPage 是跟线程相关的，push首先从tls 中取出当前线程相关连的AutoreleasePoolPage指针，类似于opengl context的原理，把POOL_BOUNDARY 放入到AutoreleasePoolPage的next位置的地方

objc_autoreleasePoolPop

C++ void objc_autoreleasePoolPop(void *ctxt) {     AutoreleasePoolPage::pop(ctxt); }     void AutoreleasePoolPage::pop(void *token)     {         AutoreleasePoolPage *page;         id *stop;                return popPage<false>(token, page, stop);     }          template<bool allowDebug>     static void     popPage(void *token, AutoreleasePoolPage *page, id *stop)     {         page->releaseUntil(stop);      }     //stop 最接近的一个POOL_BOUNDARY 的地址     void releaseUntil(id *stop)     {         // Not recursive: we don't want to blow out the stack         // if a thread accumulates a stupendous amount of garbage                 while (this->next != stop) {             // Restart from hotPage() every time, in case -release             // autoreleased more objects             AutoreleasePoolPage *page = hotPage();             id obj = *--page->next;             if (obj != POOL_BOUNDARY) {                 objc_release(obj);             }         }     }     void     objc_release(id obj)     {         if (obj->isTaggedPointerOrNil()) return;         return obj->release();     }                 inline void         objc_object::release()         {             ASSERT(!isTaggedPointer());                     rootRelease(true, RRVariant::FastOrMsgSend);         }                         ALWAYS_INLINE bool         objc_object::rootRelease()         {             return rootRelease(true, RRVariant::Fast);         }         bool         objc_object::rootRelease(bool performDealloc, objc_object::RRVariant variant)         {                          // don't check newisa.fast_rr; we already called any RR overrides             uintptr_t carry;             newisa.bits = subc(newisa.bits, RC_ONE, 0, &carry);  // extra_rc--             //#define slowpath(x) (__builtin_expect(bool(x), 0))             //    __builtin_expect((x),1)表示 x 的值为真的可能性更大；             //bool isDeallocating() {             //  return extra_rc == 0;                 if (slowpath(newisa.isDeallocating()))                 goto deallocate;              return false;          deallocate:            if (performDealloc) {                 ((void(*)(objc_object *, SEL))objc_msgSend)(this, @selector(dealloc));             }          }

Autorelease 源码解析

[pObjecdt autorelease];

也就是看下这个里边到底在干什么

C++ // Replaced by ObjectAlloc - (id)autorelease {     return _objc_rootAutorelease(self); }     id _objc_rootAutorelease(id obj)     {         ASSERT(obj);         return obj->rootAutorelease();     }         inline id objc_object::rootAutorelease()         {             if (isTaggedPointer()) return (id)this;                         return rootAutorelease2();         }             id objc_object::rootAutorelease2()             {                 ASSERT(!isTaggedPointer());                 return AutoreleasePoolPage::autorelease((id)this);             }                             static inline id AutoreleasePoolPage::autorelease(id obj)                 {                     id *dest __unused = autoreleaseFast(obj);                     return obj;                 }                     static inline id *AutoreleasePoolPage::autoreleaseFast(id obj)                     {                       //这里就跟PoolPush 方法调用逻辑是一样的了，就把这个指针入到page页面里面                     }

经过上边分析，我们看下边代码情景会生发生什么

Case 1:

Objective-C     NSObject *pObjecdt = nil;     @autoreleasepool     {         pObjecdt = [[Test alloc] init];         [pObjecdt autorelease];         NSLog(@"refcount = %d", CFGetRetainCount(pObjecdt));     }     NSLog(@"refcount = %d", CFGetRetainCount(pObjecdt));

Case 2:

Objective-C     @autoreleasepool {         NSObject *pObjecdt1 = [[[Test alloc] init] retain];         @autoreleasepool         {             NSObject *pObjecdt = [[Test alloc] init];             [pObjecdt autorelease];             _objc_autoreleasePoolPrint();         }     }

相关源码

Objective-C #ifndef Test_h #define Test_h @interface Test : NSObject {     } -(void)dealloc; + (NSString *)description; @end #endif /* Test_h */

Objective-C #import <Foundation/Foundation.h> #import "Test.h" @implementation Test -(void)dealloc {     NSLog(@"Test Destory!"); } + (NSString *)description {     return @"Test"; } @end

总结：

也就是auto release pool 只有调用对象调用 auto release才会起作用，如果没有，可以忽略这个auto release pool