python源码笔记_python源码学习笔记（二）

(二) python 继承和多态

这非常类似C++的功能，只不过是是在C基础上开发的。由上一节知，python的所有对象的基础都是PyObject，所以例如创建一个PyIntObject对象，是通过PyObejct*变量来维护，所以在python内部各个函数之间传递的都是一种范型指针PyObject* ，是不是很像C++里面的基类。如果要Print(PyIntObject* )，由多态(polymophism)我们会知道，调用的实际上是PyIntObject对象对应的类型对象中定义的输出操作。

看代码：

longPyObject_Hash(PyObject*v) //注意是PyObject

{

PyTypeObject*tp = v->ob_type; //找到类型

if (tp->tp_hash !=NULL)return (*tp->tp_hash)(v); //调用相应类型的hash函数

/*To keep to the general practice that inheriting

* solely from object in C code should work without

* an explicit call to PyType_Ready, we implicitly call

* PyType_Ready here and then check the tp_hash slot again*/

//为了维持在C代码单继承中不直接调用PyType_Ready这一惯例，在这里间接地调用PyType_Ready()，并再次检查tp_hash槽

if (tp->tp_dict ==NULL) {if (PyType_Ready(tp) < 0)return -1;if (tp->tp_hash !=NULL)return (*tp->tp_hash)(v);

}if (tp->tp_compare == NULL && RICHCOMPARE(tp) ==NULL) {return _Py_HashPointer(v); /*Use address as hash value*/ //把地址作为hash值返回

}/*If there's a cmp but no hash defined, the object can't be hashed*/

//如果有cmp，但是hash没有被定义，返回这个对象不能被hash

returnPyObject_HashNotImplemented(v);

}

以PyIntObject为例，观察其实现过程。

1 [intobject.h]2 typedef struct{3 PyObject_HEAD4 longob_ival;5 } PyIntObject;6

7 [intobject.c]8 static PyObject * //注意这里是静态函数，而且是PyObject的指针，这个是多态的典型特征

9 int_add(PyIntObject *v, PyIntObject *w) //加

10 {11 register longa, b, x;12 CONVERT_TO_LONG(v, a);13 CONVERT_TO_LONG(w, b);14 /*casts in the line below avoid undefined behaviour on overflow*/

15 x = (long)((unsigned long)a +b);16 if ((x^a) >= 0 || (x^b) >= 0)17 returnPyInt_FromLong(x);18 return PyLong_Type.tp_as_number->nb_add((PyObject *)v, (PyObject *)w);19 }20

21 static PyObject *

22 int_sub(PyIntObject *v, PyIntObject *w) //减

23 {24 register longa, b, x;25 CONVERT_TO_LONG(v, a);26 CONVERT_TO_LONG(w, b);27 /*casts in the line below avoid undefined behaviour on overflow*/

28 x = (long)((unsigned long)a -b);29 if ((x^a) >= 0 || (x^~b) >= 0)30 returnPyInt_FromLong(x);31 return PyLong_Type.tp_as_number->nb_subtract((PyObject *)v,32 (PyObject *)w);33 }34

35 static PyObject *

36 int_mul(PyObject *v, PyObject *w) //乘

37 {38 longa, b;39 long longprod; /*a*b in native long arithmetic*/

40 double doubled_longprod; /*(double)longprod*/

41 double doubleprod; /*(double)a * (double)b*/

42

43 CONVERT_TO_LONG(v, a);44 CONVERT_TO_LONG(w, b);45 /*casts in the next line avoid undefined behaviour on overflow*/

46 longprod = (long)((unsigned long)a *b);47 doubleprod = (double)a * (double)b;48 doubled_longprod = (double)longprod;49

50 /*Fast path for normal case: small multiplicands, and no info51 is lost in either method.*/

52 if (doubled_longprod ==doubleprod)53 returnPyInt_FromLong(longprod);54

55 /*Somebody somewhere lost info. Close enough, or way off? Note56 that a != 0 and b != 0 (else doubled_longprod == doubleprod == 0).57 The difference either is or isn't significant compared to the58 true value (of which doubleprod is a good approximation).59 */

60 {61 const double diff = doubled_longprod -doubleprod;62 const double absdiff = diff >= 0.0 ? diff : -diff;63 const double absprod = doubleprod >= 0.0 ?doubleprod :64 -doubleprod;65 /*absdiff/absprod <= 1/32 iff66 32 * absdiff <= absprod -- 5 good bits is "close enough"*/

67 if (32.0 * absdiff <=absprod)68 returnPyInt_FromLong(longprod);69 else

70 return PyLong_Type.tp_as_number->nb_multiply(v, w);71 }72 }73

由此可知，python的int 实际上是C里面的long实现，所以加减乘除都是用long实现，又由于PyIntObject为一个Immutable对象，这个对象不可改变，因此在最后return的都是新的对象：PyInt_FromLong(num)，即由long变量创建一个int变量。

(三)python整数的实现

在整数里，python分为大整数和小整数，为了加快计算，节省内存的分配时间，因为不论是什么对象，只要在堆上申请空间是非常费时的，所以在涉及到频繁的内存操作时需要做一些优化。python提供了一种比较原始的方法——设个阈值，无语了，有这么来的么，最起码来个动态阈值也好啊...

1 #ifndef NSMALLPOSINTS2 #define NSMALLPOSINTS 257

3 #endif

4 #ifndef NSMALLNEGINTS5 #define NSMALLNEGINTS 5

6 #endif 范围设定到(-5~257) 在这个区间里面都为小整数

7 #if NSMALLNEGINTS + NSMALLPOSINTS > 0

8 /*References to small integers are saved in this array so that they //保存在数组里面被共享9 can be shared.10 The integers that are saved are those in the range11 -NSMALLNEGINTS (inclusive) to NSMALLPOSINTS (not inclusive). //[-5, 257)12 */

13 static PyIntObject *small_ints[NSMALLNEGINTS + NSMALLPOSINTS]; //申请(NSMALLNEGINTS + NSMALLPOSINTS)个PyIntObject* 为以后所共享

14 #endif

15

然后是大整数，大整数采用块内存区间内缓存

1 #define BLOCK_SIZE 1000 /* 1K less typical malloc overhead */

2 #define BHEAD_SIZE 8 /* Enough for a 64-bit pointer */

3 #define N_INTOBJECTS ((BLOCK_SIZE - BHEAD_SIZE) / sizeof(PyIntObject))

4

5 struct_intblock {6 struct _intblock *next;7 PyIntObject objects[N_INTOBJECTS];8 };9

10 typedef struct_intblock PyIntBlock;11

12 static PyIntBlock *block_list =NULL;13 static PyIntObject *free_list = NULL;

N_INTOBJECTS    到底是多少呢，算一下PyIntObject的大小，PyIntObject宏展开(without Py_TRACE_REFS)后就是：

Py_ssize_t  ob_refnt;

PyTypeObject *ob_type;

long ob_ival;

字节大小为4+4+8 = 16 , N_INTOBJECTS  = (1000-8)/16 = 82，即一个PyIntBlock维护着的82个PyIntObeject，咋一看其实就是个单链表，因此这个82个objects相当于是82个数。

通过block_list 来维护，看代码：

1 static PyIntObject *

2 fill_free_list(void)3 {4 PyIntObject *p, *q;5 /*Python's object allocator isn't appropriate for large blocks.*/

6 p = (PyIntObject *) PyMem_MALLOC(sizeof(PyIntBlock));7 if (p ==NULL)8 return (PyIntObject *) PyErr_NoMemory();9 ((PyIntBlock *)p)->next = block_list; //串联blocklist

10 block_list = (PyIntBlock *)p;11 /*Link the int objects together, from rear to front, then return12 the address of the last int object in the block.*/

13 p = &((PyIntBlock *)p)->objects[0]; //第一块即front，头部，注意是地址哦，有个“&”

14 q = p + N_INTOBJECTS; //从0开始计数，q不是尾指针，还要减去1才是

15 while (--q > p) // //最后--q 后即rear，尾部从后往前遍历

16 Py_TYPE(q) = (struct _typeobject *)(q-1); //将所有的PyIntObject 串起来

17 Py_TYPE(q) = NULL; //q现在为头指针，类型置为空

18 return p + N_INTOBJECTS - 1; //返回rear

19 }

由上可知，相当于对外是一个blocklist, 对内是一系列的PyIntObject，当需要重新开辟blocklist时，通过((PyIntBlock *)p)->next = block_list把这些链表串起来。

现在我们可以看看PyInt_FromLong的实现了

1 [intobject.c]2 PyObject *

3 PyInt_FromLong(longival)4 {5 register PyIntObject *v; //用寄存器操作，加快速度

6 #if NSMALLNEGINTS + NSMALLPOSINTS > 0

7 if (-NSMALLNEGINTS <= ival && ival < NSMALLPOSINTS) { //在小数的范围区间，直接命中

8 v = small_ints[ival +NSMALLNEGINTS];9 Py_INCREF(v); //引用计数加1

10 #ifdef COUNT_ALLOCS //这是要统计了

11 if (ival >= 0)12 quick_int_allocs++; //正数命中的个数

13 else

14 quick_neg_int_allocs++; //负数命中的个数

15 #endif

16 return (PyObject *) v;17 }18 #endif

19 if (free_list == NULL) { //如果是大数，且free_list 没有被赋值，开始创建

20 if ((free_list = fill_free_list()) == NULL) //这里我们知道freelist指向的是链表的rear21 //和 block_list 是指向PyIntBlock 的指针相区别

22 returnNULL;23 }24 /*Inline PyObject_New*/

25 v =free_list;26 free_list = (PyIntObject *)Py_TYPE(v); //强制转换一下类型变成PyIntObject类型

27 PyObject_INIT(v, &PyInt_Type); //初始化为python的int类型

28 v->ob_ival =ival;29 return (PyObject *) v;30 }31

最后有

1 [intobject.c]2 #define PyInt_CheckExact(op) ((op)->ob_type == &PyInt_Type)

3 ...4 static void

5 int_dealloc(PyIntObject *v)6 {7 if(PyInt_CheckExact(v)) {8 Py_TYPE(v) = (struct _typeobject *)free_list; //如果是整数类对象，只是简单的把v置成free_list，即空闲链表的起点，相当于9 //覆盖的形式

10 free_list =v;11 }12 else

13 Py_TYPE(v)->tp_free((PyObject *)v); //如果不是整数类型，调用底层的释放函数

14 }15

16 static void

17 int_free(PyIntObject *v)18 {19 Py_TYPE(v) = (struct _typeobject *)free_list; //同上

20 free_list =v;21 }

不要搞混的是，上述实现不管是小数还是大数都是起着缓冲池的作用，不要误解为是实现大数字功能，只不过都是用long实现的，这个很容易误导。

再看小整数的换冲池：

1 static PyIntObject *small_ints[NSMALLNEGINTS + NSMALLPOSINTS]; //声明为静态指针数组

2 .........3

4 int

5 _PyInt_Init(void)6 {7 PyIntObject *v;8 intival;9 #if NSMALLNEGINTS + NSMALLPOSINTS > 0

10 for (ival = -NSMALLNEGINTS; ival < NSMALLPOSINTS; ival++) {11 if (!free_list && (free_list = fill_free_list()) == NULL) //同样是申请空闲链表

12 return 0;13 /*PyObject_New is inlined*/

14 v =free_list;15 free_list = (PyIntObject *)Py_TYPE(v);16 PyObject_INIT(v, &PyInt_Type);17 v->ob_ival = ival; //赋值后加入small_ints这个缓冲池

18 small_ints[ival + NSMALLNEGINTS] = v; //相当于一个一一映射的关系

19 }20 #endif

21 return 1;22 }

我们知道的就是说所有的整数都在堆里面有内存，小整数使用通过small_ints[]数组一一映射加快查找速度，大数则需要通过链表来维护内存。