1. 程式人生 > >Python3 原始碼閱讀 - 垃圾回收機制

Python3 原始碼閱讀 - 垃圾回收機制

Python的垃圾回收機制包括了兩大部分: - **引用計數**(大部分在 `Include/object.h` 中定義) - **標記清除+隔代回收**(大部分在 `Modules/gcmodule.c` 中定義) ## 1. 引用計數機制 python中萬物皆物件,他的核心結構是:`PyObject` ```c++ typedef __int64 ssize_t; typedef ssize_t Py_ssize_t; typedef struct _object { _PyObject_HEAD_EXTRA Py_ssize_t ob_refcnt; // Py_ssize_t __int64 struct _typeobject *ob_type; } PyObject; typedef struct { PyObject ob_base; Py_ssize_t ob_size; /* Number of items in variable part */ } PyVarObject; ``` `PyObject`是每個物件的底層資料結構,其中`ob_refcnt`就是作為引用計數。當一個物件有新的引用時, 它的`ob_refcnt`就會增加,當引用它的物件被刪除,它的`ob_refcnt`就會減少,當引用技術為0時,該物件的生命結束了。 1. 引用計數+1的情況 - 物件被建立 eg: a=2 - 物件被引用 eg: b=a - 物件被作為引數,傳入到一個函式中,例如func(a) - 物件作為一個元素,儲存在容器中,例如list1=[a, b] 2. 引用計數-1的情況 - 物件的別名被顯示的銷燬 eg: del a - 物件的別名被賦予新的物件 eg: a=34 - 一個物件離開它的作用域, 例如f函式執行完畢時,func函式中的區域性變數(全域性變數不會) - 物件所在的容器被銷燬,或者從容器中刪除 **如何檢視物件的引用計數** ```python import sys a = 'hello' sys.getrefcount(a) // 注意: getrefcount(a) 傳入a時, a的引用計數會加1 ``` ### 1.1 什麼時候觸發回收 當一個物件的引用計數變為了 0, 會直接進入釋放空間的流程 ```c++ /* cpython/Include/object.h */ static inline void _Py_DECREF(const char *filename, int lineno, PyObject *op) { _Py_DEC_REFTOTAL; if (--op->ob_refcnt != 0) { #ifdef Py_REF_DEBUG if (op->ob_refcnt < 0) { _Py_NegativeRefcount(filename, lineno, op); } #endif } else { /* // _Py_Dealloc 會找到對應型別的 descructor, 並且呼叫這個 descructor destructor dealloc = Py_TYPE(op)->tp_dealloc; (*dealloc)(op); */ _Py_Dealloc(op); } } ``` ## 2. 常駐記憶體物件 引用計數機制所帶來的維護引用計數的額外操作,與python執行中所進行的記憶體分配、釋放、引用賦值的次數是成正比的,這一點,相對於主流的垃圾回收技術,比如標記--清除`(mark--sweep)`、停止--複製`(stop--copy)`等方法相比是一個弱點,因為它們帶來額外操作只和記憶體數量有關,至於多少人引用了這塊記憶體則不關心。因此為了與引用計數搭配、在記憶體的分配和釋放上獲得最高的效率,python設計了大量的記憶體池機制,比如**小整數物件池、字串的intern機制,列表的freelist緩衝池**等等,這些大量使用的面向特定物件的記憶體池機制正是為了彌補引用計數的軟肋。 ### 2.1 小整數物件池 ```c++ #ifndef NSMALLPOSINTS #define NSMALLPOSINTS 257 #endif #ifndef NSMALLNEGINTS #define NSMALLNEGINTS 5 #endif #if NSMALLNEGINTS + NSMALLPOSINTS >
0 /* Small integers are preallocated in this array so that they can be shared. The integers that are preallocated are those in the range -NSMALLNEGINTS (inclusive) to NSMALLPOSINTS (not inclusive). */ static PyLongObject small_ints[NSMALLNEGINTS + NSMALLPOSINTS]; Py_INCREF(op) 增加物件引用計數 Py_DECREF(op) 減少物件引用計數, 如果計數位0, 呼叫_Py_Dealloc _Py_Dealloc(op) 呼叫對應型別的 tp_dealloc 方法 ``` 小整數物件池就是一個`PyLongObject` 陣列, 大小=257+5=262, 範圍是`[-5, 257)` 注意左閉右開. >
python對小整數的定義是[-5, 257), 這些整數物件是提前建立好的,不會被垃圾回收,在一個python程式中,所有位於這個範圍內的整數使用的都是同一個物件 ### 2.2 大整數物件池 疑惑:《Python原始碼剖析》提到的整數物件池block_list應該已經不存在了(因為PyLongObject為變長物件)。`Python2`中的`PyIntObject`實際是對`c`中的`long`的包裝。所以`Python2`也提供了專門的快取池,供大整數輪流使用,避免每次使用不斷的`malloc`分配記憶體帶來的效率損耗,可參考[劉志軍老師的講解](https://foofish.net/python_int_implement.html)。既然沒有池了,malloc/free會帶來的不小效能損耗。Guido認為Py3.0有極大的優化空間,在字串和整形操作上可以取得很好的優化結果。 ```c /* Allocate a new int object with size digits. Return NULL and set exception if we run out of memory. */ #define MAX_LONG_DIGITS \ ((PY_SSIZE_T_MAX - offsetof(PyLongObject, ob_digit))/sizeof(digit)) PyLongObject * _PyLong_New(Py_ssize_t size) { PyLongObject *result; /* Number of bytes needed is: offsetof(PyLongObject, ob_digit) + sizeof(digit)*size. Previous incarnations of this code used sizeof(PyVarObject) instead of the offsetof, but this risks being incorrect in the presence of padding between the PyVarObject header and the digits. */ if (size >
(Py_ssize_t)MAX_LONG_DIGITS) { PyErr_SetString(PyExc_OverflowError, "too many digits in integer"); return NULL; } result = PyObject_MALLOC(offsetof(PyLongObject, ob_digit) + size*sizeof(digit)); if (!result) { PyErr_NoMemory(); return NULL; } return (PyLongObject*)PyObject_INIT_VAR(result, &PyLong_Type, size); } ``` > `result = PyObject_MALLOC(offsetof(PyLongObject, ob_digit) + size*sizeof(digit));` 每一個大整數,均建立一個新的物件。id(num)均不同。 ### 2.4 字串的intern機制 > ``` > Objects/unicodeobject.c > Objects/codeobject.c > ``` > `PyStringObject`物件的intern機制之目的是:對於被intern之後的字串,比如“Ruby”,在整個Python的執行期間,系統中都只有唯一的一個與字串“Ruby”對應的`PyStringObject`物件。這樣當判斷兩個`PyStringObject`物件是否相同時,如果它們都被intern了,那麼只需要簡單地檢查它們對應的`PyObject*`是否相同即可。這個機制既節省了空間,又簡化了對`PyStringObject`物件的比較,嗯,可謂是一箭雙鵰哇。 > > 摘自:《Python原始碼剖析》 — 陳儒 Python3中`PyUnicodeObject`物件的`intern`機制和Python2的`PyStringObject`物件`intern`機制一樣,主要為了節省記憶體的開銷,利用字串物件的不可變性,對存在的字串物件重複利用 ```powershell In [50]: a = 'python' In [51]: b = 'python' In [52]: id(a) Out[52]: 442782398256 In [53]: id(b) Out[53]: 442782398256 In [54]: b = 'hello python' In [55]: a = 'hello python' In [56]: id(a) Out[56]: 442808585520 In [57]: id(b) Out[57]: 442726541488 ``` **什麼樣的字串會使用`intern`機制?** intern機制跟編譯時期有關,相關程式碼在`Objects/codeobject.c` ```c++ /* Intern selected string constants */ static int intern_string_constants(PyObject *tuple) { int modified = 0; Py_ssize_t i; for (i = PyTuple_GET_SIZE(tuple); --i >= 0; ) { PyObject *v = PyTuple_GET_ITEM(tuple, i); if (PyUnicode_CheckExact(v)) { if (PyUnicode_READY(v) == -1) { PyErr_Clear(); continue; } if (all_name_chars(v)) { PyObject *w = v; PyUnicode_InternInPlace(&v); if (w != v) { PyTuple_SET_ITEM(tuple, i, v); modified = 1; } } } /*....*/ } /* all_name_chars(s): true iff s matches [a-zA-Z0-9_]* */ static int all_name_chars(PyObject *o) { const unsigned char *s, *e; if (!PyUnicode_IS_ASCII(o)) return 0; s = PyUnicode_1BYTE_DATA(o); e = s + PyUnicode_GET_LENGTH(o); for (; s != e; s++) { if (!Py_ISALNUM(*s) && *s != '_') return 0; } return 1; } ``` > 可見 all_name_chars 決定了是否會 intern,簡單來說就是 ascii 字母,數字和下劃線組成的字串會被快取。但是不僅如此。2.5還會說 ```c++ /* This dictionary holds all interned unicode strings. Note that references to strings in this dictionary are *not* counted in the string's ob_refcnt. When the interned string reaches a refcnt of 0 the string deallocation function will delete the reference from this dictionary. Another way to look at this is that to say that the actual reference count of a string is: s->ob_refcnt + (s->state ? 2 : 0) */ static PyObject *interned = NULL; /*省略*/ void PyUnicode_InternInPlace(PyObject **p) { PyObject *s = *p; PyObject *t; #ifdef Py_DEBUG assert(s != NULL); assert(_PyUnicode_CHECK(s)); #else if (s == NULL || !PyUnicode_Check(s)) return; #endif /* If it's a subclass, we don't really know what putting it in the interned dict might do. */ if (!PyUnicode_CheckExact(s)) return; // [1] if (PyUnicode_CHECK_INTERNED(s)) return; if (interned == NULL) { interned = PyDict_New(); if (interned == NULL) { PyErr_Clear(); /* Don't leave an exception */ return; } } Py_ALLOW_RECURSION // [2] t = PyDict_SetDefault(interned, s, s); Py_END_ALLOW_RECURSION if (t == NULL) { PyErr_Clear(); return; } // [3] if (t != s) { Py_INCREF(t); Py_SETREF(*p, t); return; } // [4] /* The two references in interned are not counted by refcnt. The deallocator will take care of this */ Py_REFCNT(s) -= 2; _PyUnicode_STATE(s).interned = SSTATE_INTERNED_MORTAL; } ``` 通過函式我們可以得知,python中維護這一個interned變數的指標,這個變數指向`PyDict_New`建立的物件,而`PyDict_New`實際上建立了一個`PyDictObject`物件,是Python中`dict`型別的物件。實際上intern機制就是維護一個字典,這個字典中記錄著被intern機制處理過的字串物件,`[1]`處`PyUnicode_CHECK_INTERNED`巨集檢查字串物件的`state.interned`是否被標記, 如果字串物件的`state.interned`被標記了,就直接返回;`[2]`處**嘗試**把沒有被標記的`字串物件s`作為`key-value`加入`interned`字典中;`[3]`處表示`字串物件s`已經在`interned`字典中(對應的value值是`字串物件t`),(通過`Py_SETREF`巨集來改變p指標的指向),且原`字串物件p`會因引用計數為零被回收。`Py_SETREF`巨集在`Include/object.h`定義著: ```c /* Safely decref `op` and set `op` to `op2`. * * As in case of Py_CLEAR "the obvious" code can be deadly: * * Py_DECREF(op); * op = op2; * * The safe way is: * * Py_SETREF(op, op2); * * That arranges to set `op` to `op2` _before_ decref'ing, so that any code * triggered as a side-effect of `op` getting torn down no longer believes * `op` points to a valid object. * * Py_XSETREF is a variant of Py_SETREF that uses Py_XDECREF instead of * Py_DECREF. */ #define Py_SETREF(op, op2) \ do { \ PyObject *_py_tmp = (PyObject *)(op); \ (op) = (op2); \ Py_DECREF(_py_tmp); \ } while (0) ``` `[4]`中把新加入`interned`字典中的字串物件做減引用操作,並把`state.interned`標記成`SSTATE_INTERNED_MORTAL`。`SSTATE_INTERNED_MORTAL`表示字串物件被intern機制處理,但會隨著引用計數被回收;`interned`標記還有另外一種`SSTATE_INTERNED_IMMORTAL`,表示被intern機制處理但物件不可銷燬,會與Python直譯器同在。`PyUnicode_InternInPlace`只能建立`SSTATE_INTERNED_MORTAL`狀態的字串,要想建立`SSTATE_INTERNED_IMMORTAL`狀態的字串需要通過另外一個介面,強制改變intern的狀態 ```c++ void PyUnicode_InternImmortal(PyObject **p) { PyUnicode_InternInPlace(p); if (PyUnicode_CHECK_INTERNED(*p) != SSTATE_INTERNED_IMMORTAL) { _PyUnicode_STATE(*p).interned = SSTATE_INTERNED_IMMORTAL; Py_INCREF(*p); } } ``` **為什麼引用`Py_REFCNT(s) -= 2;`要-2呢?** ```c++ PyDict_SetDefault(PyObject *d, PyObject *key, PyObject *defaultobj) { PyDictObject *mp = (PyDictObject *)d; PyObject *value; Py_hash_t hash; /*...*/ if (ix == DKIX_EMPTY) { /*...*/ Py_ssize_t hashpos = find_empty_slot(mp->ma_keys, hash); ep0 = DK_ENTRIES(mp->ma_keys); ep = &ep0[mp->ma_keys->dk_nentries]; dictkeys_set_index(mp->ma_keys, hashpos, mp->ma_keys->dk_nentries); Py_INCREF(key); Py_INCREF(value); /*...*/ return value; } ``` > 對於被intern機制處理了的PyStringObject物件,Python採用了特殊的引用計數機制。在將一個PyStringObject物件a的PyObject指標**作為key和value**新增到interned中時,PyDictObject物件會通過這兩個指標對a的引用計數進行兩次加1的操作。但是Python的設計者規定在interned中a的指標不能被視為物件a的有效引用,因為如果是有效引用的話,那麼a的引用計數在Python結束之前永遠都不可能為0,因為interned中至少有兩個指標引用了a,那麼刪除a就永遠不可能了,這顯然是沒有道理的。 > 摘自:《Python原始碼剖析》 — 陳儒 **注意:**實際上,即使Python會對一個字串進行intern機制的處理,也會先建立一個`PyUnicodeObject`物件,然後檢查在`interned`字典中是否有值和其相同,存在的話就將`interned`字典儲存的value值返回,之前臨時建立的字串物件會由於引用計數為零而回收。 **是否可以直接對C原生物件做intern的動作呢?不需要建立臨時物件** 事實上`CPython`確實提供了以`char * `為引數的intern機制相關函式,但是,也是一樣的建立temp在設定intern. ```c++ PyUnicode_InternImmortal(PyObject **p) { PyUnicode_InternInPlace(p); if (PyUnicode_CHECK_INTERNED(*p) != SSTATE_INTERNED_IMMORTAL) { _PyUnicode_STATE(*p).interned = SSTATE_INTERNED_IMMORTAL; Py_INCREF(*p); } } ``` **為什麼需要臨時物件?** > 因為PyDict_SetDefault() 操作的是PyDictObject物件,而該物件必須以PyObject*指標作為鍵 ### 2.5 字元緩衝池(單字元) python為小整數物件準備了小整數物件池,當然對於常用的字元,python對應的也建了字串緩衝池,因為 python3 中通過 `unicode_latin1[256] `**將長度為 1 的 ascii 的字元也快取了** ```c /* Single character Unicode strings in the Latin-1 range are being shared as well. */ static PyObject *unicode_latin1[256] = {NULL}; unicode_decode_utf8(){ /*省略*/ /* ASCII is equivalent to the first 128 ordinals in Unicode. */ if (size == 1 && (unsigned char)s[0] < 128) { if (consumed) *consumed = 1; return get_latin1_char((unsigned char)s[0]); } /*省略*/ } static PyObject* get_latin1_char(unsigned char ch) { PyObject *unicode = unicode_latin1[ch]; if (!unicode) { unicode = PyUnicode_New(1, ch); if (!unicode) return NULL; PyUnicode_1BYTE_DATA(unicode)[0] = ch; assert(_PyUnicode_CheckConsistency(unicode, 1)); unicode_latin1[ch] = unicode; } Py_INCREF(unicode); return unicode; } ``` ```powershell In [46]: a = 'p' In [47]: b = 'p' In [48]: id(a) Out[48]: 442757120384 In [49]: id(b) Out[49]: 442757120384 ``` **當然單字元也包括空字元。** ```c++ /* The empty Unicode object is shared to improve performance. */ static PyObject *unicode_empty = NULL; ``` ```shell In [8]: a = 'hello' + 'python' In [9]: b = 'hellopython' In [10]: a is b Out[10]: True In [11]: a = 'hello ' + 'python' In [12]: b = 'hello python' In [13]: id(a) Out[13]: 118388503536 In [14]: id(b) Out[14]: 118387544240 In [15]: 'hello ' + 'python' is 'hello python' Out[15]: False In [16]: 'hello_' + 'python' is 'hello_python' Out[16]: True ``` ### 2.6 小結: - **小整數[-5, 257)共用物件,常駐記憶體** - **單個字母,長度為 1 的 ascii 的字元[latin1](https://en.wikipedia.org/wiki/ISO/IEC_8859-1)會被interned, 包括空字元,共用物件,常駐記憶體** - **由字母、數字、下劃線([a-zA-Z0-9_])組成的字串,不可修改,預設開啟intern機制,共用物件,引用計數為0時,銷燬** - 字串(含有空格),不可修改,沒開啟intern機制,不共用物件,引用計數為0,銷燬 ## 3. 標記清除+分代回收 為了防止出現**迴圈引用**的致命性問題,**python採用的是引用計數機制為主,標記-清除和分代收集兩種機制為輔的策略**。 ![image.png](https://i.loli.net/2020/06/11/2GI7nmWSPvNjpOR.png) ![image.png](https://i.loli.net/2020/06/11/iLS5ATeBo3W1sVr.png) 我們設定 n1.next 指向 n2,同時設定 n2.prev 指回 n1,現在,我們的兩個節點使用迴圈引用的方式構成了一個`雙向連結串列`,同時請注意到 ABC 以及 DEF 的引用計數值已經增加到了2,現在,假定我們的程式不再使用這兩個節點了,我們將 n1 和 n2 都設定為None,Python會像往常一樣將每個節點的引用計數減少到1。 ![image.png](https://i.loli.net/2020/06/11/q6IpCneRGwt2iM3.png) ### 3.1 在python中的零代(Generation Zero) Ruby使用一個連結串列(free_list)來持續追蹤未使用的、自由的物件,Python使用一種不同的連結串列來持續追蹤活躍的物件。而不將其稱之為“活躍列表”,Python的內部C程式碼將其稱為零代(Generation Zero)。每次當你建立一個物件或其他什麼值的時候,Python會將其加入零代連結串列: ![image.png](https://i.loli.net/2020/06/11/Yp8UMNXmP4ORFDG.png) 從上邊可以看到當我們建立ABC節點的時候,Python將其加入零代連結串列。請注意到這並不是一個真正的列表,並不能直接在你的程式碼中訪問,事實上這個連結串列是一個完全內部的Python執行時。 ***疑惑1:*****對於容器物件(比如list、dict、class、instance等等),是在什麼時候繫結GC,放入第0連結串列呢?** 相似的,當我們建立DEF節點的時候,Python將其加入同樣的連結串列: ![image.png](https://i.loli.net/2020/06/11/giHxS7pYNKwT8kh.png) 現在零代包含了兩個節點物件。(他還將包含Python建立的每個其他值,與一些Python自己使用的內部值。) ### 3.2 標記迴圈引用 當達到某個 閾值之後 直譯器會迴圈遍歷,迴圈遍歷零代列表上的每個物件,檢查列表中每個互相引用的物件,根據規則減掉其引用計數。在這個過程中,Python會一個接一個的統計內部引用的數量以防過早地釋放物件。以下例子便於理解: ![image.png](https://i.loli.net/2020/06/11/Y1ORJmN2UxPnFg7.png) 從上面可以看到 ABC 和 DEF 節點包含的引用數為1.有三個其他的物件同時存在於零代連結串列中,藍色的箭頭指示了有一些物件正在被零代連結串列之外的其他物件所引用。 ![image.png](https://i.loli.net/2020/06/11/g1fsLIiYMZupBRo.png) 通過識別內部引用,Python能夠減少許多零代連結串列物件的引用計數。在上圖的第一行中你能夠看見ABC和DEF的引用計數已經變為零了,這意味著收集器可以釋放它們並回收記憶體空間了。剩下的活躍的物件則被移動到一個新的連結串列:一代連結串列。 **疑惑2: 內部如何識別零代的迴圈引用計數,在什麼閾值下會觸發GC執行?** ### 3.3 在原始碼中摸索答案 Python通過`PyGC_Head`來跟蹤container物件,`PyGC_Head`資訊位於`PyObject_HEAD`之前,定義在`Include/objimpl.h`中 ```c++ typedef union _gc_head { struct { union _gc_head *gc_next; union _gc_head *gc_prev; Py_ssize_t gc_refs; } gc; double dummy; /* force worst-case alignment */ } PyGC_Head; ``` **表頭資料結構** ```c++ //Include/internal/mem.h struct gc_generation { PyGC_Head head; int threshold; /* collection threshold */ // 閾值 int count; /* count of allocations or collections of younger generations */ // 實時個數 }; ``` Python中用於分代垃圾收集的三個“代”由`_gc_runtime_state.generations`陣列所表示著: **解答疑惑2,三個代的閾值如下陣列** ```c++ /* If we change this, we need to cbhange the default value in the signature of gc.collect. */ #define NUM_GENERATIONS 3 _PyGC_Initialize(struct _gc_runtime_state *state) { state->enabled = 1; /* automatic collection enabled? */ #define _GEN_HEAD(n) (&state->generations[n].head) struct gc_generation generations[NUM_GENERATIONS] = { /* PyGC_Head, threshold, count */ {{{_GEN_HEAD(0), _GEN_HEAD(0), 0}}, 700, 0}, {{{_GEN_HEAD(1), _GEN_HEAD(1), 0}}, 10, 0}, {{{_GEN_HEAD(2), _GEN_HEAD(2), 0}}, 10, 0}, }; for (int i = 0; i < NUM_GENERATIONS; i++) { state->generations[i] = generations[i]; }; state->generation0 = GEN_HEAD(0); struct gc_generation permanent_generation = { {{&state->permanent_generation.head, &state->permanent_generation.head, 0}}, 0, 0 }; state->permanent_generation = permanent_generation; } ``` ![image.png](https://i.loli.net/2020/06/11/bkKPog8n3Cy1TOx.png) **解答疑惑1:那container物件是什麼時候加入第0“代”的container物件連結串列呢?** 對於python內建物件的建立,container物件是通過`PyObject_GC_New`函式來建立的,而非container物件是通過`PyObject_Malloc`函式來建立的。 ```c++ // Include/objimpl.h #define PyObject_GC_New(type, typeobj) \ ( (type *) _PyObject_GC_New(typeobj) ) // 呼叫了Modules/gcmodule.c中的_PyObject_GC_New函式: PyObject * _PyObject_GC_New(PyTypeObject *tp) { PyObject *op = _PyObject_GC_Malloc(_PyObject_SIZE(tp)); if (op != NULL) op = PyObject_INIT(op, tp); return op; } static PyObject * _PyObject_GC_Alloc(int use_calloc, size_t basicsize) { PyObject *op; PyGC_Head *g; size_t size; if (basicsize > PY_SSIZE_T_MAX - sizeof(PyGC_Head)) return PyErr_NoMemory(); size = sizeof(PyGC_Head) + basicsize; // [1] 申請PyGC_Head和物件本身的記憶體 if (use_calloc) g = (PyGC_Head *)PyObject_Calloc(1, size); else g = (PyGC_Head *)PyObject_Malloc(size); if (g == NULL) return PyErr_NoMemory(); // [2] 設定gc_refs的值 g->gc.gc_refs = 0; _PyGCHead_SET_REFS(g, GC_UNTRACKED); // [3] generations[0].count++; /* number of allocated GC objects */ if (generations[0].count > generations[0].threshold && enabled && generations[0].threshold && !collecting && !PyErr_Occurred()) { collecting = 1; collect_generations(); collecting = 0; } // [4] FROM_GC巨集定義可以通過PyGC_Head地址轉換PyObject_HEAD地址,逆運算是AS_GC巨集定義。 op = FROM_GC(g); return op; } PyObject * _PyObject_GC_Malloc(size_t basicsize) { return _PyObject_GC_Alloc(0, basicsize); } ``` [4] `FROM_GC`巨集定義可以通過`PyGC_Head`地址轉換`PyObject_HEAD`地址,逆運算是`AS_GC`巨集定義。 ```c++ /* Get an object's GC head */ #define AS_GC(o) ((PyGC_Head *)(o)-1) /* Get the object given the GC head */ #define FROM_GC(g) ((PyObject *)(((PyGC_Head *)g)+1)) ``` **當觸發閾值後,是如何進行GC回收的?** `collect`是垃圾回收的主入口函式。**特別注意 finalizers 與 python 的`__del__`綁定了**。 ```c /* This is the main function. Read this to understand how the * collection process works. */ static Py_ssize_t collect(int generation, Py_ssize_t *n_collected, Py_ssize_t *n_uncollectable, int nofail) { int i; Py_ssize_t m = 0; /* # objects collected */ Py_ssize_t n = 0; /* # unreachable objects that couldn't be collected */ PyGC_Head *young; /* the generation we are examining */ PyGC_Head *old; /* next older generation */ PyGC_Head unreachable; /* non-problematic unreachable trash */ PyGC_Head finalizers; /* objects with, & reachable from, __del__ */ PyGC_Head *gc; _PyTime_t t1 = 0; /* initialize to prevent a compiler warning */ struct gc_generation_stats *stats = &_PyRuntime.gc.generation_stats[generation]; ... // “標記-清除”前的準備 // 垃圾標記 // 垃圾清除 ... /* Update stats */ if (n_collected) *n_collected = m; if (n_uncollectable) *n_uncollectable = n; stats->collections++; stats->collected += m; stats->uncollectable += n; if (PyDTrace_GC_DONE_ENABLED()) PyDTrace_GC_DONE(n+m); return n+m; } ``` ### 3.3.1 標記-清除前的準備 ```c // [1] /* update collection and allocation counters */ if (generation+1 < NUM_GENERATIONS) _PyRuntime.gc.generations[generation+1].count += 1; for (i = 0; i <= generation; i++) _PyRuntime.gc.generations[i].count = 0; // [2] /* merge younger generations with one we are currently collecting */ for (i = 0; i < generation; i++) { gc_list_merge(GEN_HEAD(i), GEN_HEAD(generation)); } // [3] /* handy references */ young = GEN_HEAD(generation); if (generation < NUM_GENERATIONS-1) old = GEN_HEAD(generation+1); else old = young; // [4] /* Using ob_refcnt and gc_refs, calculate which objects in the * container set are reachable from outside the set (i.e., have a * refcount greater than 0 when all the references within the * set are taken into account). */ update_refs(young); subtract_refs(young); ``` [1] 先更新了將被回收的“代”以及老一“代”的count計數器。 這邊對老一“代”的count計數器增量1就可以看出來在第1“代”和第2“代”的count值其實表示的是該代垃圾回收的次數。 [2] 通過`gc_list_merge`函式將這些“代”合併成一個連結串列。 ``` /* append list `from` onto list `to`; `from` becomes an empty list */ static void gc_list_merge(PyGC_Head *from, PyGC_Head *to) { PyGC_Head *tail; assert(from != to); if (!gc_list_is_empty(from)) { tail = to->gc.gc_prev; tail->gc.gc_next = from->gc.gc_next; tail->gc.gc_next->gc.gc_prev = tail; to->gc.gc_prev = from->gc.gc_prev; to->gc.gc_prev->gc.gc_next = to; } gc_list_init(from); } static void gc_list_init(PyGC_Head *list) { list->gc.gc_prev = list; list->gc.gc_next = list; } ``` `gc_list_merge`函式將from連結串列連結到to連結串列末尾並把from連結串列置為空連結串列。 [3] 經過合併操作之後,所有需要被進行垃圾回收的物件都連結到young“代”(滿足超過閾值的最老“代”),並記錄old“代”,後面需要將不可回收的物件移到old“代”。 連結串列的合併操作: ![image.png](https://i.loli.net/2020/06/11/g4QiLXuzaD2eKEo.png) [4] 尋找root object集合 要對合並的連結串列進行垃圾標記,首先需要尋找root object集合。 所謂的root object即是一些全域性引用和函式棧中的引用。這些引用所用的物件是不可被刪除的。 ```python list1 = [] list2 = [] list1.append(list2) list2.append(list1) a = list1 del list1 del list2 ``` 上面的Python中迴圈引用的程式碼,變數a所指向的物件就是root object。 三色標記模型 ### 3.3.2 垃圾標記 ```c // [1] /* Leave everything reachable from outside young in young, and move * everything else (in young) to unreachable. * NOTE: This used to move the reachable objects into a reachable * set instead. But most things usually turn out to be reachable, * so it's more efficient to move the unreachable things. */ gc_list_init(&unreachable); move_unreachable(young, &unreachable); // [2] /* Move reachable objects to next generation. */ if (young != old) { if (generation == NUM_GENERATIONS - 2) { _PyRuntime.gc.long_lived_pending += gc_list_size(young); } gc_list_merge(young, old); } else { /* We only untrack dicts in full collections, to avoid quadratic dict build-up. See issue #14775. */ untrack_dicts(young); _PyRuntime.gc.long_lived_pending = 0; _PyRuntime.gc.long_lived_total = gc_list_size(young); } ``` [1] 初始化不可達連結串列,呼叫`move_unreachable`函式將迴圈引用的物件移動到不可達連結串列中: ```c /* Move the unreachable objects from young to unreachable. After this, * all objects in young have gc_refs = GC_REACHABLE, and all objects in * unreachable have gc_refs = GC_TENTATIVELY_UNREACHABLE. All tracked * gc objects not in young or unreachable still have gc_refs = GC_REACHABLE. * All objects in young after this are directly or indirectly reachable * from outside the original young; and all objects in unreachable are * not. */ static void move_unreachable(PyGC_Head *young, PyGC_Head *unreachable) { PyGC_Head *gc = young->gc.gc_next; /* Invariants: all objects "to the left" of us in young have gc_refs * = GC_REACHABLE, and are indeed reachable (directly or indirectly) * from outside the young list as it was at entry. All other objects * from the original young "to the left" of us are in unreachable now, * and have gc_refs = GC_TENTATIVELY_UNREACHABLE. All objects to the * left of us in 'young' now have been scanned, and no objects here * or to the right have been scanned yet. */ while (gc != young) { PyGC_Head *next; if (_PyGCHead_REFS(gc)) { /* gc is definitely reachable from outside the * original 'young'. Mark it as such, and traverse * its pointers to find any other objects that may * be directly reachable from it. Note that the * call to tp_traverse may append objects to young, * so we have to wait until it returns to determine * the next object to visit. */ PyObject *op = FROM_GC(gc); traverseproc traverse = Py_TYPE(op)->tp_traverse; assert(_PyGCHead_REFS(gc) > 0); _PyGCHead_SET_REFS(gc, GC_REACHABLE); (void) traverse(op, (visitproc)visit_reachable, (void *)young); next = gc->gc.gc_next; if (PyTuple_CheckExact(op)) { _PyTuple_MaybeUntrack(op); } } else { /* This *may* be unreachable. To make progress, * assume it is. gc isn't directly reachable from * any object we've already traversed, but may be * reachable from an object we haven't gotten to yet. * visit_reachable will eventually move gc back into * young if that's so, and we'll see it again. */ next = gc->gc.gc_next; gc_list_move(gc, unreachable); _PyGCHead_SET_REFS(gc, GC_TENTATIVELY_UNREACHABLE); } gc = next; } } ``` 這邊遍歷young“代”的container物件連結串列,`_PyGCHead_REFS(gc)`判斷是不是root object或從某個root object能直接/間接引用的物件,由於root object集合中的物件是不能回收的,因此,被這些物件直接或間接引用的物件也是不能回收的。 `_PyGCHead_REFS(gc)`為0並不能斷定這個物件是可回收的,但是還是先移動到`unreachable`連結串列中,設定了`GC_TENTATIVELY_UNREACHABLE`標誌表示暫且認為是不可達的,如果是存在被root object直接或間接引用的物件,這樣的物件還會被移出`unreachable`連結串列中。 [2] 將可達的物件移到下一“代”。 ### 3.3.3 垃圾清除 ```c // [1] /* All objects in unreachable are trash, but objects reachable from * legacy finalizers (e.g. tp_del) can't safely be deleted. */ gc_list_init(&finalizers); move_legacy_finalizers(&unreachable, &finalizers); /* finalizers contains the unreachable objects with a legacy finalizer; * unreachable objects reachable *from* those are also uncollectable, * and we move those into the finalizers list too. */ move_legacy_finalizer_reachable(&finalizers); // [2] /* Collect statistics on collectable objects found and print * debugging information. */ for (gc = unreachable.gc.gc_next; gc != &unreachable; gc = gc->gc.gc_next) { m++; } // [3] /* Clear weakrefs and invoke callbacks as necessary. */ m += handle_weakrefs(&unreachable, old); // [4] /* Call tp_finalize on objects which have one. */ finalize_garbage(&unreachable); // [5] if (check_garbage(&unreachable)) { revive_garbage(&unreachable); gc_list_merge(&unreachable, old); } else { /* Call tp_clear on objects in the unreachable set. This will cause * the reference cycles to be broken. It may also cause some objects * in finalizers to be freed. */ delete_garbage(&unreachable, old); } // [6] /* Collect statistics on uncollectable objects found and print * debugging information. */ for (gc = finalizers.gc.gc_next; gc != &finalizers; gc = gc->gc.gc_next) { n++; } ... // [7] /* Append instances in the uncollectable set to a Python * reachable list of garbage. The programmer has to deal with * this if they insist on creating this type of structure. */ (void)handle_legacy_finalizers(&finalizers, old); /* Clear free list only during the collection of the highest * generation */ if (generation == NUM_GENERATIONS-1) { clear_freelists(); } ``` [1] 處理`unreachable`連結串列中有finalizer的物件。即python中 實現了`__del__`魔法方法的物件 ```c /* Move the objects in unreachable with tp_del slots into `finalizers`. * Objects moved into `finalizers` have gc_refs set to GC_REACHABLE; the * objects remaining in unreachable are left at GC_TENTATIVELY_UNREACHABLE. */ static void move_legacy_finalizers(PyGC_Head *unreachable, PyGC_Head *finalizers) { PyGC_Head *gc; PyGC_Head *next; /* March over unreachable. Move objects with finalizers into * `finalizers`. */ for (gc = unreachable->gc.gc_next; gc != unreachable; gc = next) { PyObject *op = FROM_GC(gc); assert(IS_TENTATIVELY_UNREACHABLE(op)); next = gc->gc.gc_next; if (has_legacy_finalizer(op)) { gc_list_move(gc, finalizers); _PyGCHead_SET_REFS(gc, GC_REACHABLE); } } } ``` 遍歷`unreachable`連結串列,將擁有finalizer的例項物件移到`finalizers`連結串列中,並標示為`GC_REACHABLE`。 ```c /* Return true if object has a pre-PEP 442 finalization method. */ static int has_legacy_finalizer(PyObject *op) { return op->ob_type->tp_del != NULL; } ``` **擁有finalizer的例項物件指的就是實現了`tp_del`函式的物件。** ```c /* Move objects that are reachable from finalizers, from the unreachable set * into finalizers set. */ static void move_legacy_finalizer_reachable(PyGC_Head *finalizers) { traverseproc traverse; PyGC_Head *gc = finalizers->gc.gc_next; for (; gc != finalizers; gc = gc->gc.gc_next) { /* Note that the finalizers list may grow during this. */ traverse = Py_TYPE(FROM_GC(gc))->tp_traverse; (void) traverse(FROM_GC(gc), (visitproc)visit_move, (void *)finalizers); } } ``` 對`finalizers`連結串列中擁有finalizer的例項物件遍歷其引用物件,呼叫`visit_move`訪問者,這些被引用的物件也不應該被釋放。 ```c /* A traversal callback for move_legacy_finalizer_reachable. */ static int visit_move(PyObject *op, PyGC_Head *tolist) { if (PyObject_IS_GC(op)) { if (IS_TENTATIVELY_UNREACHABLE(op)) { PyGC_Head *gc = AS_GC(op); gc_list_move(gc, tolist); _PyGCHead_SET_REFS(gc, GC_REACHABLE); } } return 0; } #define IS_TENTATIVELY_UNREACHABLE(o) ( \ _PyGC_REFS(o) == GC_TENTATIVELY_UNREACHABLE) ``` `visit_move`函式將引用物件還在`unreachable`連結串列的物件移到`finalizers`連結串列中。 [2] 統計`unreachable`連結串列數量。 [3] 處理弱引用。 [4] [5] 開始清除垃圾物件,我們先只看`delete_garbage`函式: ```c /* Break reference cycles by clearing the containers involved. This is * tricky business as the lists can be changing and we don't know which * objects may be freed. It is possible I screwed something up here. */ static void delete_garbage(PyGC_Head *collectable, PyGC_Head *old) { inquiry clear; while (!gc_list_is_empty(collectable)) { PyGC_Head *gc = collectable->gc.gc_next; PyObject *op = FROM_GC(gc); if (_PyRuntime.gc.debug & DEBUG_SAVEALL) { PyList_Append(_PyRuntime.gc.garbage, op); } else { if ((clear = Py_TYPE(op)->tp_clear) != NULL) { Py_INCREF(op); clear(op); Py_DECREF(op); } } if (collectable->gc.gc_next == gc) { /* object is still alive, move it, it may die later */ gc_list_move(gc, old); _PyGCHead_SET_REFS(gc, GC_REACHABLE); } } } ``` 遍歷`unreachable`連結串列中的container物件,呼叫其型別物件的`tp_clear`指標指向的函式,我們以list物件為例: ```c static int _list_clear(PyListObject *a) { Py_ssize_t i; PyObject **item = a->ob_item; if (item != NULL) { /* Because XDECREF can recursively invoke operations on this list, we make it empty first. */ i = Py_SIZE(a); Py_SIZE(a) = 0; a->ob_item = NULL; a->allocated = 0; while (--i >= 0) { Py_XDECREF(item[i]); } PyMem_FREE(item); } /* Never fails; the return value can be ignored. Note that there is no guarantee that the list is actually empty at this point, because XDECREF may have populated it again! */ return 0; } ``` `_list_clear`函式對container物件的每個元素進行引用數減量操作並釋放container物件記憶體。 `delete_garbage`在對container物件進行`clear`操作之後,還會檢查是否成功,如果該container物件沒有從`unreachable`連結串列上摘除,表示container物件還不能銷燬,需要放回到老一“代”中,並標記`GC_REACHABLE`。 [6] 統計`finalizers`連結串列數量。 [7] 處理`finalizers`連結串列的物件。 ```c /* Handle uncollectable garbage (cycles with tp_del slots, and stuff reachable * only from such cycles). * If DEBUG_SAVEALL, all objects in finalizers are appended to the module * garbage list (a Python list), else only the objects in finalizers with * __del__ methods are appended to garbage. All objects in finalizers are * merged into the old list regardless. * Returns 0 if all OK, <0 on error (out of memory to grow the garbage list). * The finalizers list is made empty on a successful return. */ static int handle_legacy_finalizers(PyGC_Head *finalizers, PyGC_Head *old) { PyGC_Head *gc = finalizers->gc.gc_next; if (_PyRuntime.gc.garbage == NULL) { _PyRuntime.gc.garbage = PyList_New(0); if (_PyRuntime.gc.garbage == NULL) Py_FatalError("gc couldn't create gc.garbage list"); } for (; gc != finalizers; gc = gc->gc.gc_next) { PyObject *op = FROM_GC(gc); if ((_PyRuntime.gc.debug & DEBUG_SAVEALL) || has_legacy_finalizer(op)) { if (PyList_Append(_PyRuntime.gc.garbage, op) < 0) return -1; } } gc_list_merge(finalizers, old); return 0; } ``` 遍歷`finalizers`連結串列,將擁有finalizer的例項物件放到一個名為garbage的PyListObject物件中,可以通過gc模組檢視。 ```powershell >>> import gc >>> gc.garbage ``` 並把`finalizers`連結串列晉升到老一“代”。 > **注意:`__del__`給gc帶來的影響, gc模組唯一處理不了的是迴圈引用的類都有`__del__`方法,所以專案中要避免定義`__del__`方法** ### 3.4 小結 1. GC的流程: ``` -> 發現超過閾值了 -> 觸發垃圾回收 -> 將所有可達物件連結串列放到一起 -> 遍歷, 計算有效引用計數 -> 分成 有效引用計數=0 和 有效引用計數 > 0 兩個集合 -> 大於0的, 放入到更老一代 -> =0的, 執行回收 -> 回收遍歷容器內的各個元素, 減掉對應元素引用計數(破掉迴圈引用) -> 執行-1的邏輯, 若發現物件引用計數=0, 觸發記憶體回收 -> 由python底層記憶體管理機制回收記憶體 ``` 2. 觸發GC的條件 - 主動呼叫`gc.collect(),` - 當gc模組的計數器達到閥值的時候 - 程式退出的時候 ## 4. GC閾值 **分代回收 以空間換時間** > **重要思想**:將系統中的所有記憶體塊根據其存活的時間劃分為不同的集合, 每個集合就成為一個”代”, 垃圾收集的頻率隨著”代”的存活時間的增大而減小(活得越長的物件, 就越不可能是垃圾, 就應該減少去收集的頻率) **弱代假說** 分代垃圾回收演算法的核心行為:垃圾回收器會更頻繁的處理新物件。一個新的物件即是你的程式剛剛建立的,而一個來的物件則是經過了幾個時間週期之後仍然存在的物件。Python會在當一個物件從零代移動到一代,或是從一代移動到二代的過程中提升`(promote)`這個物件。 **為什麼要這麼做?**這種演算法的根源來自於弱代假說(**weak generational hypothesis**)。這個假說由兩個觀點構成: > 首先是年親的物件通常死得也快,而老物件則很有可能存活更長的時間。 假定我們建立了一個Python建立: ```python n1 = Node("ABC") ``` 根據假說,我的程式碼很可能僅僅會使用ABC很短的時間。這個物件也許僅僅只是一個方法中的中間結果,並且隨著方法的返回這個物件就將變成垃圾了。大部分的新物件都是如此般地很快變成垃圾。然而,偶爾程式會建立一些很重要的,存活時間比較長的物件-例如web應用中的session變數或是配置項。 > 通過頻繁的處理零代連結串列中的新物件,Python的垃圾收集器將把時間花在更有意義的地方:它處理那些很快就可能變成垃圾的新物件。同時只在很少的時候,當滿足閾值的條件,收集器才回去處理那些老變數。 ## 5. Python中的gc模組使用 > gc模組預設是開啟自動回收垃圾的,`gc.isenabled()=True` **常用函式:** - `gc.set_debug(flags)` 設定gc的debug日誌,一般設定為`gc.DEBUG_LEAK` ```python """ DEBUG_STATS - 在垃圾收集過程中列印所有統計資訊 DEBUG_COLLECTABLE - 打印發現的可收集物件 DEBUG_UNCOLLECTABLE - 列印unreachable物件(除了uncollectable物件) DEBUG_SAVEALL - 將物件儲存到gc.garbage(一個列表)裡面,而不是釋放它 DEBUG_LEAK - 對記憶體洩漏的程式進行debug (everything but STATS). """ ``` - `gc.collect([generation])` 顯式進行垃圾回收,可以輸入引數,0代表只檢查第一代的物件,1代表檢查一,二代的物件,2代表檢查一,二,三代的物件,如果不傳引數,執行一個full collection,也就是等於傳2。 返回不可達(unreachable objects)物件的數目 - `gc.get_threshold() `獲取的gc模組中自動執行垃圾回收的頻率 - `gc.get_stats()`檢視每一代的具體資訊 - `gc.set_threshold(threshold0[, threshold1[, threshold2])` 設定自動執行垃圾回收的頻率 - `gc.get_count() `獲取當前自動執行垃圾回收的計數器,返回一個長度為3的列表 例如**(488,3,0)**,其中488是指距離上一次一代垃圾檢查,Python分配記憶體的數目減去釋放記憶體的數目,注意是記憶體分配,而不是引用計數的增加。 3是指距離上一次二代垃圾檢查,一代垃圾檢查的次數,同理,0是指距離上一次三代垃圾檢查,二代垃圾檢查的次數。 **計數器和閾值關係解釋:** ```python 當計數器從(699,3,0)增加到(700,3,0),gc模組就會執行gc.collect(0),即檢查一代物件的垃圾,並重置計數器為(0,4,0) 當計數器從(699,9,0)增加到(700,9,0),gc模組就會執行gc.collect(1),即檢查一、二代物件的垃圾,並重置計數器為(0,0,1) 當計數器從(699,9,9)增加到(700,9,9),gc模組就會執行gc.collect(2),即檢查一、二、三代物件的垃圾,並重置計數器為(0,0,0) ``` ## 6. 工作中如何避免迴圈引用? > To avoid circular references in your code, you can use weak references, that are implemented in the `weakref` module. Unlike the usual references, the `weakref.ref` doesn't increase the reference count and returns `None` if an object was destroyed. [rushter](https://rushter.com/blog/python-garbage-collector/) ```python import weakref class Node(): def __init__(self, value): self.value = value self._parent = None self.children = [] def __repr__(self): return 'Node({!r:})'.format(self.value) @property def parent(self): return None if self._parent is None else self._parent() @parent.setter def parent(self, node): self._parent = weakref.ref(node) def add_child(self, child): self.children.append(child) child.parent = self if __name__ == '__main__': a = Data() del a a = Node() del a a = Node() a.add_child(Node()) del a ``` > 弱引用消除了引用迴圈的這個問題,本質來講,**弱引用就是一個物件指標,它不會增加它的引用計數** 為了訪問弱引用所引用的物件,你可以像函式一樣去呼叫它即可。如果那個物件還存在就會返回它,否則就返回一個None。 由於原始物件的引用計數沒有增加,那麼就可以去刪除它了 **參考文章和書籍:** 1. [visualizing garbage collection in ruby and python](http://patshaughnessy.net/2013/10/24/visualizing-garbage-collection-in-ruby-and-python) 2. [膜拜的大佬-Junnplus'blog](https://github.com/Junnplus/blog/issues/19) 3. [wklken前輩](http://wklken.me/posts/2015/09/29/python-source-gc.html) 4. [The Garbage Collector](https://pythoninternal.wordpress.com/2014/08/04/the-garbage-collector/) 5. [Garbage collection in Python: things you need to know](https://rushter.com/blog/python-garbage-collector/) 6. [Python-CookBook-迴圈引用資料結構的記憶體管理](https://python3-cookbook.readthedocs.io/zh_CN/latest/c08/p23_managing_memory_in_cyclic_data_structures.html) 7. 《python原始碼剖析》 8. Python-3.8.3/Modules/gcm