python
yield
的功能相似於return
,可是不一樣之處在於它返回的是生成器
。app
生成器是經過一個或多個yield
表達式構成的函數,每個生成器都是一個迭代器(可是迭代器不必定是生成器)。函數
若是一個函數包含yield
關鍵字,這個函數就會變爲一個生成器。oop
生成器並不會一次返回全部結果,而是每次遇到yield
關鍵字後返回相應結果,並保留函數當前的運行狀態,等待下一次的調用。源碼分析
因爲生成器也是一個迭代器,那麼它就應該支持next
方法來獲取下一個值。性能
# 經過`yield`來建立生成器 def func(): for i in xrange(10); yield i # 經過列表來建立生成器 [i for i in xrange(10)]
# 調用以下 >>> f = func() >>> f # 此時生成器尚未運行 <generator object func at 0x7fe01a853820> >>> f.next() # 當i=0時,遇到yield關鍵字,直接返回 0 >>> f.next() # 繼續上一次執行的位置,進入下一層循環 1 ... >>> f.next() 9 >>> f.next() # 當執行完最後一次循環後,結束yield語句,生成StopIteration異常 Traceback (most recent call last): File "<stdin>", line 1, in <module> StopIteration >>>
除了next
函數,生成器還支持send
函數。該函數能夠向生成器傳遞參數。ui
>>> def func(): ... n = 0 ... while 1: ... n = yield n #能夠經過send函數向n賦值 ... >>> f = func() >>> f.next() # 默認狀況下n爲0 0 >>> f.send(1) #n賦值1 1 >>> f.send(2) 2 >>>
最經典的例子,生成無限序列。this
常規的解決方法是,生成一個知足要求的很大的列表,這個列表須要保存在內存中,很明顯內存限制了這個問題。lua
def get_primes(start): for element in magical_infinite_range(start): if is_prime(element): return element
若是使用生成器就不須要返回整個列表,每次都只是返回一個數據,避免了內存的限制問題。code
def get_primes(number): while True: if is_prime(number): yield number number += 1
生成器的源碼在Objects/genobject.c
。
在解釋生成器以前,須要講解一下Python虛擬機的調用原理。
Python虛擬機有一個棧幀的調用棧,其中棧幀的是PyFrameObject
,位於Include/frameobject.h
。
typedef struct _frame { PyObject_VAR_HEAD struct _frame *f_back; /* previous frame, or NULL */ PyCodeObject *f_code; /* code segment */ PyObject *f_builtins; /* builtin symbol table (PyDictObject) */ PyObject *f_globals; /* global symbol table (PyDictObject) */ PyObject *f_locals; /* local symbol table (any mapping) */ PyObject **f_valuestack; /* points after the last local */ /* Next free slot in f_valuestack. Frame creation sets to f_valuestack. Frame evaluation usually NULLs it, but a frame that yields sets it to the current stack top. */ PyObject **f_stacktop; PyObject *f_trace; /* Trace function */ /* If an exception is raised in this frame, the next three are used to * record the exception info (if any) originally in the thread state. See * comments before set_exc_info() -- it's not obvious. * Invariant: if _type is NULL, then so are _value and _traceback. * Desired invariant: all three are NULL, or all three are non-NULL. That * one isn't currently true, but "should be". */ PyObject *f_exc_type, *f_exc_value, *f_exc_traceback; PyThreadState *f_tstate; int f_lasti; /* Last instruction if called */ /* Call PyFrame_GetLineNumber() instead of reading this field directly. As of 2.3 f_lineno is only valid when tracing is active (i.e. when f_trace is set). At other times we use PyCode_Addr2Line to calculate the line from the current bytecode index. */ int f_lineno; /* Current line number */ int f_iblock; /* index in f_blockstack */ PyTryBlock f_blockstack[CO_MAXBLOCKS]; /* for try and loop blocks */ PyObject *f_localsplus[1]; /* locals+stack, dynamically sized */ } PyFrameObject;
棧幀保存了給出代碼的的信息和上下文,其中包含最後執行的指令,全局和局部命名空間,異常狀態等信息。f_valueblock
保存了數據,b_blockstack
保存了異常和循環控制方法。
舉一個例子來講明,
def foo(): x = 1 def bar(y): z = y + 2 # <--- (3) ... and the interpreter is here. return z return bar(x) # <--- (2) ... which is returning a call to bar ... foo() # <--- (1) We're in the middle of a call to foo ...
那麼,相應的調用棧以下,一個py文件,一個類,一個函數都是一個代碼塊,對應者一個Frame,保存着上下文環境以及字節碼指令。
c --------------------------- a | bar Frame | -> block stack: [] l | (newest) | -> data stack: [1, 2] l --------------------------- | foo Frame | -> block stack: [] s | | -> data stack: [<Function foo.<locals>.bar at 0x10d389680>, 1] t --------------------------- a | main (module) Frame | -> block stack: [] c | (oldest) | -> data stack: [<Function foo at 0x10d3540e0>] k ---------------------------
每個棧幀都擁有本身的數據棧和block棧,獨立的數據棧和block棧使得解釋器能夠中斷和恢復棧幀(生成器正式利用這點)。
Python代碼首先被編譯爲字節碼,再由Python虛擬機來執行。通常來講,一條Python語句對應着多條字節碼(因爲每條字節碼對應着一條C語句,而不是一個機器指令,因此不能按照字節碼的數量來判斷代碼性能)。
調用dis
模塊能夠分析字節碼,
from dis import dis dis(foo) 5 0 LOAD_CONST 1 (1) # 加載常量1 3 STORE_FAST 0 (x) # x賦值爲1 6 6 LOAD_CONST 2 (<code object bar at 0x7f3cdee3a030, file "t1.py", line 6>) # 加載常量2 9 MAKE_FUNCTION 0 # 建立函數 12 STORE_FAST 1 (bar) 9 15 LOAD_FAST 1 (bar) 18 LOAD_FAST 0 (x) 21 CALL_FUNCTION 1 # 調用函數 24 RETURN_VALUE
其中,
第一行爲代碼行號; 第二行爲偏移地址; 第三行爲字節碼指令; 第四行爲指令參數; 第五行爲參數解釋。
由了上面對於調用棧的理解,就能夠很容易的明白生成器的具體實現。
生成器的源碼位於object/genobject.c
。
PyObject * PyGen_New(PyFrameObject *f) { PyGenObject *gen = PyObject_GC_New(PyGenObject, &PyGen_Type); # 建立生成器對象 if (gen == NULL) { Py_DECREF(f); return NULL; } gen->gi_frame = f; # 賦予代碼塊 Py_INCREF(f->f_code); # 引用計數+1 gen->gi_code = (PyObject *)(f->f_code); gen->gi_running = 0; # 0表示爲執行,也就是生成器的初始狀態 gen->gi_weakreflist = NULL; _PyObject_GC_TRACK(gen); # GC跟蹤 return (PyObject *)gen; }
next
與send
函數,以下
static PyObject * gen_iternext(PyGenObject *gen) { return gen_send_ex(gen, NULL, 0); } static PyObject * gen_send(PyGenObject *gen, PyObject *arg) { return gen_send_ex(gen, arg, 0); }
從上面的代碼中能夠看到,send
和next
都是調用的同一函數gen_send_ex
,區別在因而否帶有參數。
static PyObject * gen_send_ex(PyGenObject *gen, PyObject *arg, int exc) { PyThreadState *tstate = PyThreadState_GET(); PyFrameObject *f = gen->gi_frame; PyObject *result; if (gen->gi_running) { # 判斷生成器是否已經運行 PyErr_SetString(PyExc_ValueError, "generator already executing"); return NULL; } if (f==NULL || f->f_stacktop == NULL) { # 若是代碼塊爲空或調用棧爲空,則拋出StopIteration異常 /* Only set exception if called from send() */ if (arg && !exc) PyErr_SetNone(PyExc_StopIteration); return NULL; } if (f->f_lasti == -1) { # f_lasti=1 表明首次執行 if (arg && arg != Py_None) { # 首次執行不容許帶有參數 PyErr_SetString(PyExc_TypeError, "can't send non-None value to a " "just-started generator"); return NULL; } } else { /* Push arg onto the frame's value stack */ result = arg ? arg : Py_None; Py_INCREF(result); # 該參數引用計數+1 *(f->f_stacktop++) = result; # 參數壓棧 } /* Generators always return to their most recent caller, not * necessarily their creator. */ f->f_tstate = tstate; Py_XINCREF(tstate->frame); assert(f->f_back == NULL); f->f_back = tstate->frame; gen->gi_running = 1; # 修改生成器執行狀態 result = PyEval_EvalFrameEx(f, exc); # 執行字節碼 gen->gi_running = 0; # 恢復爲未執行狀態 /* Don't keep the reference to f_back any longer than necessary. It * may keep a chain of frames alive or it could create a reference * cycle. */ assert(f->f_back == tstate->frame); Py_CLEAR(f->f_back); /* Clear the borrowed reference to the thread state */ f->f_tstate = NULL; /* If the generator just returned (as opposed to yielding), signal * that the generator is exhausted. */ if (result == Py_None && f->f_stacktop == NULL) { Py_DECREF(result); result = NULL; /* Set exception if not called by gen_iternext() */ if (arg) PyErr_SetNone(PyExc_StopIteration); } if (!result || f->f_stacktop == NULL) { /* generator can't be rerun, so release the frame */ Py_DECREF(f); gen->gi_frame = NULL; } return result; }
PyEval_EvalFrameEx
函數的功能爲執行字節碼並返回結果。
# 主要流程以下, for (;;) { switch(opcode) { # opcode爲操做碼,對應着各類操做 case NOP: goto fast_next_opcode; ... ... case YIELD_VALUE: # 若是操做碼是yield retval = POP(); f->f_stacktop = stack_pointer; why = WHY_YIELD; goto fast_yield; # 利用goto跳出循環 } } fast_yield: ... return vetval; # 返回結果
舉一個例子,f_back
上一個Frame,f_lasti
上一次執行的指令的偏移量,
import sys from dis import dis def func(): f = sys._getframe(0) print f.f_lasti print f.f_back yield 1 print f.f_lasti print f.f_back yield 2 a = func() dis(func) a.next() a.next()
結果以下,其中第三行的英文爲操做碼,對應着上面的opcode
,每次switch都是在不一樣的opcode
之間進行選擇。
6 0 LOAD_GLOBAL 0 (sys) 3 LOAD_ATTR 1 (_getframe) 6 LOAD_CONST 1 (0) 9 CALL_FUNCTION 1 12 STORE_FAST 0 (f) 7 15 LOAD_FAST 0 (f) 18 LOAD_ATTR 2 (f_lasti) 21 PRINT_ITEM 22 PRINT_NEWLINE 8 23 LOAD_FAST 0 (f) 26 LOAD_ATTR 3 (f_back) 29 PRINT_ITEM 30 PRINT_NEWLINE 9 31 LOAD_CONST 2 (1) 34 YIELD_VALUE # 此時操做碼爲YIELD_VALUE,直接跳轉上述goto語句,此時f_lasti爲當前指令,f_back爲當前frame 35 POP_TOP 11 36 LOAD_FAST 0 (f) 39 LOAD_ATTR 2 (f_lasti) 42 PRINT_ITEM 43 PRINT_NEWLINE 12 44 LOAD_FAST 0 (f) 47 LOAD_ATTR 3 (f_back) 50 PRINT_ITEM 51 PRINT_NEWLINE 13 52 LOAD_CONST 3 (2) 55 YIELD_VALUE 56 POP_TOP 57 LOAD_CONST 0 (None) 60 RETURN_VALUE 18 <frame object at 0x7fa75fcebc20> #和下面的frame相同,屬於同一個frame,也就是說在同一個函數(命名空間)內,frame是同一個。 39 <frame object at 0x7fa75fcebc20>