RabbitMQ架构

typedef struct {
    long hash;      /* cached hash code for the entry key */
    PyObject *key;
} setentry;
/*
This data structure is shared by set and frozenset objects.
*/
typedef struct _setobject PySetObject;
struct _setobject {
    PyObject_HEAD
    Py_ssize_t fill;  /* # Active + # Dummy */
    Py_ssize_t used;  /* # Active */
    /* The table contains mask + 1 slots, and that's a power of 2.
     * We store the mask instead of the size because the mask is more
     * frequently needed.
     */
    Py_ssize_t mask;
    /* table points to smalltable for small tables, else to
     * additional malloc'ed memory.  table is never NULL!  This rule
     * saves repeated runtime null-tests.
     */
    setentry *table;
    setentry *(*lookup)(PySetObject *so, PyObject *key, long hash);
    setentry smalltable[PySet_MINSIZE];
    long hash;                  /* only used by frozenset objects */
    PyObject *weakreflist;      /* List of weak references */
};

setentry是哈希表中的元素，记录插入元素的哈希值以及对应的Python对象。PySetObject是哈希表的具体结构：

fill 被填充的键的个数，包括Active和dummy，稍后解释具体意思
used 被填充的键中有效的个数，即集合中的元素个数
mask 哈希表的长度的掩码，数值为容量值减一
table 存放元素的数组的指针
smalltable 默认的存放元素的数组

当元素较少时，所有元素只存放在smalltable数组中，此时table指向smalltable。当元素增多，会从新分配内存存放所有的元素，此时smalltable没有用，table指向新分配的内存。

哈希表中的元素有三种状态：

active 元素有效，此时setentry.key != null && != dummy
dummy 元素无效key=dummy，此插槽(slot)存放的元素已经被删除
NULL 无元素，此插槽从来没有被使用过

dummy是为了表明当前位置存放过元素，需要继续查找。假设a和b元素具有相同的哈希值，所以b只能放在冲撞函数指向的第二个位置。先删除a，再去查找b。如果a被设置为NULL，那么无法确定b是不存在还是应该继续探查第二个位置，所以a只能被设置为dummy。查找b的过程中，第一个位置为dummy所以继续探查，直到找到b；或者直到NULL，证明b确实不存在。

Set中的缓存

set中会存在缓存系统，缓存数量为80个_setobject结构。

/* Reuse scheme to save calls to malloc, free, and memset */
#ifndef PySet_MAXFREELIST
#define PySet_MAXFREELIST 80
#endif
static PySetObject *free_list[PySet_MAXFREELIST];
static int numfree = 0;
static void
set_dealloc(PySetObject *so)
{
    register setentry *entry;
    Py_ssize_t fill = so->fill;
    PyObject_GC_UnTrack(so);
    Py_TRASHCAN_SAFE_BEGIN(so)
    if (so->weakreflist != NULL)
        PyObject_ClearWeakRefs((PyObject *) so);
    // 释放每个setentry
    for (entry = so->table; fill > 0; entry++) {
        if (entry->key) {
            --fill;
            Py_DECREF(entry->key);
        }
    }
    // 如果分配了内存存放setentry，则释放掉
    if (so->table != so->smalltable)
        PyMem_DEL(so->table);
    // 缓存_setobject
    if (numfree < PySet_MAXFREELIST && PyAnySet_CheckExact(so))
        free_list[numfree++] = so;
    else
        Py_TYPE(so)->tp_free(so);
    Py_TRASHCAN_SAFE_END(so)
}
}

freelist缓存只会对_setobject结构本身起效，会释放掉额外分配的存储键的内存。

Set中查找元素

set中元素查找有两个函数，在默认情况下的查找函数为set_lookkey_string。当发现查找的元素不是string类型时，会将对应的lookup函数设置为set_lookkey，然后调用该函数。

static setentry *
set_lookkey_string(PySetObject *so, PyObject *key, register long hash)
{
    register Py_ssize_t i;
    register size_t perturb;
    register setentry *freeslot;
    register size_t mask = so->mask;
    setentry *table = so->table;
    register setentry *entry;
    /* Make sure this function doesn't have to handle non-string keys,
       including subclasses of str; e.g., one reason to subclass
       strings is to override __eq__, and for speed we don't cater to
       that here. */
       
    /*
    * 元素不是string，设置lookup = set_lookkey并调用
    */
    if (!PyString_CheckExact(key)) {
        so->lookup = set_lookkey;
        return set_lookkey(so, key, hash);
    }
    // 元素是字符串
    i = hash & mask;
    entry = &table[i];
    // 插槽为空，或者插槽上的key的内存地址与被查找一致
    if (entry->key == NULL || entry->key == key)
        return entry;
    // 第一个插槽为dummy，需要继续调用冲撞函数查找
    if (entry->key == dummy)
        freeslot = entry;
    // 第一个插槽为其他元素，检查是否相等
    else {
        if (entry->hash == hash && _PyString_Eq(entry->key, key))
            return entry;
        freeslot = NULL;
    }
    /* In the loop, key == dummy is by far (factor of 100s) the
       least likely outcome, so test for that last. */
    /* 第一个插槽为dummy，继续查找 */
    for (perturb = hash; ; perturb >>= PERTURB_SHIFT) {
        // 冲撞函数
        i = (i << 2) + i + perturb + 1;
        entry = &table[i & mask];
        if (entry->key == NULL)
            return freeslot == NULL ? entry : freeslot;
        if (entry->key == key
            || (entry->hash == hash
            && entry->key != dummy
            && _PyString_Eq(entry->key, key)))
            return entry;
        // 记录第一个为dummy的插槽，当key不存在是返回该插槽
        if (entry->key == dummy && freeslot == NULL)
            freeslot = entry;
    }
    assert(0);          /* NOT REACHED */
    return 0;
}

查找函数最后返回的插槽有三种情况：

key存在，返回此插槽
key不存在，对应的插槽为NULL，返回此插槽
key不存在，对应的插槽有dummy，返回第一个dummy的插槽

set_lookkey与此类似，只不过比较元素时需要调用对应的比较函数。

set的重新散列

为了减少哈希冲撞，当哈希表中的元素数量太多时需要扩大桶的长度以减少冲撞。Python中当填充的元素大于总的2/3时开始重新散列，会重新分配一个有效元素个数的两倍或者四倍的新的散列表。

static int
set_add_key(register PySetObject *so, PyObject *key)
{
    register long hash;
    register Py_ssize_t n_used;
    if (!PyString_CheckExact(key) ||
        (hash = ((PyStringObject *) key)->ob_shash) == -1) {
        hash = PyObject_Hash(key);
        if (hash == -1)
            return -1;
    }
    assert(so->fill <= so->mask);  /* at least one empty slot */
    n_used = so->used;
    Py_INCREF(key);
    if (set_insert_key(so, key, hash) == -1) {
        Py_DECREF(key);
        return -1;
    }
    // 填充的元素 > 2/3 总数量
    if (!(so->used > n_used && so->fill*3 >= (so->mask+1)*2))
        return 0;
    // 新分配的内存为2倍或者4倍有效元素的个数。
    // 可以知道一般情况下，有效元素占新分配元素的 1/6
    // 再占满一半才需要再次分配(2/3 - 1/6 = 1/2)
    return set_table_resize(so, so->used>50000 ? so->used*2 : so->used*4);
}

(完）

Python源码剖析—信号处理机制

发表于 2016-10-23 | 分类于 python源码剖析 |

Python信号处理机制

本篇的信号处理机制不是指Python的signal模块的使用，而是指Python解释器本身如何处理信号以及如何实现signal模块。Python解释器处理信号机制需要做好两件事情：

Python解释器与操作系统有关信号的交互
Python解释器实现信号语义的API接口和模块

大体上，Python解释器对信号的实现总体思路比较简单。Python解释器对信号做一层封装，在这层封装中处理信号，以及信号发生时的回调函数，使之能够纳入整个Python虚拟机的运行中。我们先从信号的初始化开始一点点揭露整个运作机制。

信号机制的初始化

信号机制的初始化是在Python初始化整个解释器时开始的，Python在初始化函数中调用initsigs来进行整个系统以及singal模块的初始化。

// python/pythonrun.c  
void
Py_InitializedEx(int install_sigs)
{
        ....
    
        if (install_sigs)   // 初始化时install_sigs==1
        initsigs(); /* Signal handling stuff, including initintr() */
        
        ....
}
// 代码中PyOS_xxx系列都是Python解释器直接对系统调用的封装
static void
initsigs(void)
{
#ifdef SIGPIPE
    PyOS_setsig(SIGPIPE, SIG_IGN);     // 忽略SIGPIPE  
#endif
#ifdef SIGXFZ
    PyOS_setsig(SIGXFZ, SIG_IGN);      // 忽略SIGXFZ 
#endif
#ifdef SIGXFSZ
    PyOS_setsig(SIGXFSZ, SIG_IGN);     // 忽略SIGXFSZ  file size exceeded
#endif
    PyOS_InitInterrupts(); /* May imply initsignal() */
}
// python/modules/singalmodle.c
void
PyOS_InitInterrupts(void)
{
    initsignal();
    _PyImport_FixupExtension("signal", "signal");
}

直接进入到singalmodule.c中看signal模块以及信号的初始化。

PyMODINIT_FUNC
initsignal(void)
{
    PyObject *m, *d, *x;
    int i;
#ifdef WITH_THREAD
    main_thread = PyThread_get_thread_ident();
    main_pid = getpid();
#endif
    /* Create the module and add the functions */
    // 初始化signal模块
    m = Py_InitModule3("signal", signal_methods, module_doc);
    if (m == NULL)
        return;
    /* Add some symbolic constants to the module */
    d = PyModule_GetDict(m);
    // 将SIG_DFL、SIGIGN 转化成Python整数对象
    x = DefaultHandler = PyLong_FromVoidPtr((void *)SIG_DFL);
    if (!x || PyDict_SetItemString(d, "SIG_DFL", x) < 0)
        goto finally;
    x = IgnoreHandler = PyLong_FromVoidPtr((void *)SIG_IGN);
    if (!x || PyDict_SetItemString(d, "SIG_IGN", x) < 0)
        goto finally;
    x = PyInt_FromLong((long)NSIG);
    if (!x || PyDict_SetItemString(d, "NSIG", x) < 0)
        goto finally;
    Py_DECREF(x);
    /*
    * 获取signal模块中的默认中断处理函数，
    * 实际就是 signal_default_int_handler    
    */
    x = IntHandler = PyDict_GetItemString(d, "default_int_handler");
    if (!x)
        goto finally;
    Py_INCREF(IntHandler);
    /*
    * 初始化Python解释器中的Handler，
    * 这个数组存储每个用户自定义的信号处理函数
    * 以及标志是否发生该信号的标志。
    */
    Handlers[0].tripped = 0;
    for (i = 1; i < NSIG; i++) {
        void (*t)(int);
        t = PyOS_getsig(i);
        Handlers[i].tripped = 0;
        if (t == SIG_DFL)
            Handlers[i].func = DefaultHandler;
        else if (t == SIG_IGN)
            Handlers[i].func = IgnoreHandler;
        else
            Handlers[i].func = Py_None; /* None of our business */
        Py_INCREF(Handlers[i].func);
    }
    
    /* 
    * 为 SIGINT 设置Python解释器的信号处理函数signal_handler
    * signal_handler 也会成为Python解释器与用户自定义处理函数的桥梁
    */
    if (Handlers[SIGINT].func == DefaultHandler) {
        /* Install default int handler */
        Py_INCREF(IntHandler);
        Py_DECREF(Handlers[SIGINT].func);
        Handlers[SIGINT].func = IntHandler;
        old_siginthandler = PyOS_setsig(SIGINT, signal_handler);
    }
// 实现signal模块中的各个 SIGXXX 信号值和名称
#ifdef SIGHUP                  
    x = PyInt_FromLong(SIGHUP);
    PyDict_SetItemString(d, "SIGHUP", x);
    Py_XDECREF(x);
#endif
    ....
    if (!PyErr_Occurred())
        return;
    /* Check for errors */
  finally:
    return;
}

可以看到Python将用户自定义信号处理函数保存在Handler数组中，而实际上向系统注册signal_handler函数。这个signal_handler函数成为信号发生时沟通Python解释器和用户自定义信号处理函数的桥梁。可以从signal.signal的实现中清楚的看到这一点。

// python/signalmodule.c
static PyObject *
signal_signal(PyObject *self, PyObject *args)
{
    PyObject *obj;
    int sig_num;
    PyObject *old_handler;
    void (*func)(int);
    if (!PyArg_ParseTuple(args, "iO:signal", &sig_num, &obj))
        return NULL;
#ifdef MS_WINDOWS
    /* Validate that sig_num is one of the allowable signals */
    switch (sig_num) {
        case SIGABRT: break;
#ifdef SIGBREAK
        /* Issue #10003: SIGBREAK is not documented as permitted, but works
           and corresponds to CTRL_BREAK_EVENT. */
        case SIGBREAK: break;
#endif
        case SIGFPE: break;
        case SIGILL: break;
        case SIGINT: break;
        case SIGSEGV: break;
        case SIGTERM: break;
        default:
            PyErr_SetString(PyExc_ValueError, "invalid signal value");
            return NULL;
    }
#endif
#ifdef WITH_THREAD
    // 只有主函数才能设置信号处理函数
    if (PyThread_get_thread_ident() != main_thread) {
        PyErr_SetString(PyExc_ValueError,
                        "signal only works in main thread");
        return NULL;
    }
#endif
    if (sig_num < 1 || sig_num >= NSIG) {
        PyErr_SetString(PyExc_ValueError,
                        "signal number out of range");
        return NULL;
    }
    if (obj == IgnoreHandler)
        func = SIG_IGN;
    else if (obj == DefaultHandler)
        func = SIG_DFL;
    else if (!PyCallable_Check(obj)) {
        PyErr_SetString(PyExc_TypeError,
"signal handler must be signal.SIG_IGN, signal.SIG_DFL, or a callable object");
                return NULL;
    }
    else
        // 除了signal.SIG_IGN和signal.SIG_DFL之外
        // Python解释器向系统注册的都是signal_handler函数
        func = signal_handler;
    if (PyOS_setsig(sig_num, func) == SIG_ERR) {
        PyErr_SetFromErrno(PyExc_RuntimeError);
        return NULL;
    }
    // 把实际的用户自定义信号处理函数，放入对应的Handler数组中
    // tripped标记对应信号值的信号是否发生了
    old_handler = Handlers[sig_num].func;
    Handlers[sig_num].tripped = 0;
    Py_INCREF(obj);
    Handlers[sig_num].func = obj;
    if (old_handler != NULL)
        return old_handler;
    else
        Py_RETURN_NONE;
}

信号产生时Python的动作

当信号产生时，操作系统会调用Python解释器注册的信号处理函数，即上文中的signal_handler函数。这个函数将对应的Handler结构中的信号产生标志tripped设置为1，然后将一个统一信号处理函数trip_signal作为pending_call注册到Python虚拟机的执行栈中。于是，Python在虚拟机执行过程中调用pending_call并执行各个用户自定义的信号处理函数。

static void
signal_handler(int sig_num)
{
    int save_errno = errno;
#if defined(WITH_THREAD) && defined(WITH_PTH)
    if (PyThread_get_thread_ident() != main_thread) {
        pth_raise(*(pth_t *) main_thread, sig_num);
    }
    else
#endif
    {
#ifdef WITH_THREAD
    /* See NOTES section above */
    if (getpid() == main_pid)
#endif
    {
        trip_signal(sig_num);
    }
#ifndef HAVE_SIGACTION
#ifdef SIGCHLD
    /* To avoid infinite recursion, this signal remains
       reset until explicit re-instated.
       Don't clear the 'func' field as it is our pointer
       to the Python handler... */
    if (sig_num != SIGCHLD)
#endif
    /* If the handler was not set up with sigaction, reinstall it.  See
     * Python/pythonrun.c for the implementation of PyOS_setsig which
     * makes this true.  See also issue8354. */
    // 重新设置信号处理
    PyOS_setsig(sig_num, signal_handler);
#endif
    }
    /* Issue #10311: asynchronously executing signal handlers should not
       mutate errno under the feet of unsuspecting C code. */
    errno = save_errno;
}
static void
trip_signal(int sig_num)
{
    // 信号产生了
    Handlers[sig_num].tripped = 1;
    
    // 如果正在处理信号，则不再向Python虚拟机提交
    if (is_tripped)
        return;
    /* Set is_tripped after setting .tripped, as it gets
       cleared in PyErr_CheckSignals() before .tripped. */
    is_tripped = 1;
    // 向Python虚拟机提交pending_call，纳入到整个虚拟机的执行过程中
    Py_AddPendingCall(checksignals_witharg, NULL);
    if (wakeup_fd != -1)
        write(wakeup_fd, "\0", 1);
}
static int
checksignals_witharg(void * arg)
{
    return PyErr_CheckSignals();
}
int
PyErr_CheckSignals(void)
{
    int i;
    PyObject *f;
    // 已经在信号处理中
    if (!is_tripped)
        return 0;
    // 只主线程中处理信号
#ifdef WITH_THREAD
    if (PyThread_get_thread_ident() != main_thread)
        return 0;
#endif
    /*
     * The is_tripped variable is meant to speed up the calls to
     * PyErr_CheckSignals (both directly or via pending calls) when no
     * signal has arrived. This variable is set to 1 when a signal arrives
     * and it is set to 0 here, when we know some signals arrived. This way
     * we can run the registered handlers with no signals blocked.
     *
     * NOTE: with this approach we can have a situation where is_tripped is
     *       1 but we have no more signals to handle (Handlers[i].tripped
     *       is 0 for every signal i). This won't do us any harm (except
     *       we're gonna spent some cycles for nothing). This happens when
     *       we receive a signal i after we zero is_tripped and before we
     *       check Handlers[i].tripped.
     */
    /*
    * 恢复该信号。对于信号处理可能有两种情况：
    * 1. 在is_tripped = 0之前： 信号又发生了，则只在Handler中设置标志位，
    *    不会再次提交到pendingcall，多个信号只处理一次;
    * 2. 在is_tripped = 0之后： 信号又发生了，则会被再次提交到pendingcall
    *    每发生一次信号调用一次信号处理函数。
    */
    is_tripped = 0;
    if (!(f = (PyObject *)PyEval_GetFrame()))
        f = Py_None;
    // 按照信号值从小到大依次调用对应的信号处理函数
    for (i = 1; i < NSIG; i++) {
        if (Handlers[i].tripped) {
            PyObject *result = NULL;
            PyObject *arglist = Py_BuildValue("(iO)", i, f);
            Handlers[i].tripped = 0;
            if (arglist) {
                result = PyEval_CallObject(Handlers[i].func,
                                           arglist);
                Py_DECREF(arglist);
            }
            if (!result)
                return -1;
            Py_DECREF(result);
        }
    }
    return 0;
}

这里面的PyErr_CheckSignals函数也会被其他模块调用直接信号的处理。例如，在file.read读取文件过程中中断，Python对调用该函数进行信号处理。至此，可以看到整个信号处理的流程：

初始化signal模块，将对应的操作系统信号值、函数转化成Python对象
用户设置信号就向操作系统注册函数signal_handler，并将用户自定义信号处理函数设置到对应的Handler数组中
当信号发生时，操作系统调用signal_handler设置tripped=1，然后调用trip_signal将统一处理函数checksignals_witharg作为pendingcall注册到Python虚拟机的执行栈中。
Python虚拟机在处理pendingcall时调用checksignals_withargs，从而信号处理函数得以执行。
另外，Python其他模块可以直接调用PyErr_CheckSignals进行信号处理。

Python信号的语义

通过注释以及代码剖析可以归纳Python的信号语义：

只有主线程能够设置、捕获和处理信号
信号设置一直有效（signal_handler中会再次注册信号处理函数)
多次信号，可能会被合并处理一次
按照信号值从小到大处理

信号实例

主线程才能捕获信号


import threading                                                                 
import signal                                                                    
import time                                                                      
                                                                                 
SIG = []                                                                         
                                                                                 
def sig_handler(*args):                                                          
    SIG.append(args)                                                             
                                                                                 
signal.signal(signal.SIGUSR1, sig_handler)                                       
signal.signal(signal.SIGUSR2, sig_handler)                                       
signal.signal(signal.SIGSYS, sig_handler)                                        
                                                                                 
class MyThread(threading.Thread):                                                
    def run(self, *args):                                                        
        start = time.time()                                                      
        while True:                                                              
            if time.time() > start + 10:                                         
                break                                                            
        print 'In Thread:', SIG                                                          
                                                                                 
t = MyThread()                                                                   
t.start()                                                                        
print 'start thread:', t                                                         
t.join()                                                                         
print 'In Main:', SIG

root@ubuntu:/home/python# python test_signal_main_thread.py
start thread: <MyThread(Thread-1, started 140229890316032)>
In Main: []
In Thread: []          # [1]

[1] Python中的线程都是分离的，因此主线程很快退出。信号不能发送到主线程，因此不能被执行。

信号可能只被处理一次

import threading
import signal
import time
SIG = []
def sig_handler(*args):
    SIG.append(args)
signal.signal(signal.SIGUSR1, sig_handler)
signal.signal(signal.SIGUSR2, sig_handler)
signal.signal(signal.SIGSYS, sig_handler)
class MyThread(threading.Thread):
    def run(self, *args):
        start = time.time()
        while True:
            if time.time() > start + 10:
                break
        print 'In Thread:', SIG
t = MyThread()
t.start()
print 'start thread:', t
t.join()
print 'In Main:', SIG

# python test_signal_signo.py       # kill -10  2856; kill -10 2856; kill -12 2856
start thread: <MyThread(Thread-1, started 140351081834240)>
In Thread: []       # [1]
In Main: [(10, <frame object at 0x16e2d50>), (12, <frame object at 0x16e2d50>)] # [2]

[1] 主线程的t.join一直阻塞，因此在子线程没有退出前不能处理信号。（C语言的信号处理是可以打断堵塞信号的）
[2] 信号在有机会处理之前发生了两次信号，但是只处理了一次。

Python信号的特殊性

Python的信号语义与Linux的C语言的信号语义有一些不同。

Python信号的处理函数会一直有效；而Linux除非特殊设置否则信号处理函数默认只调用一次就被恢复
Python信号只能在主线程中设置、捕获和处理
Python信号不能打断堵塞操作(因为信号发生时子线程在运行)

(完）

Python源码剖析—字符串对象PyStringObject

发表于 2016-10-18 | 分类于 python源码剖析 |

Python字符串对象PyStringObject

Python的字符串对象是一个不可变对象，任何改变字符串字面值的操作都是重新创建一个新的字符串。

astr = 'astr'
id(astr)
Out[22]: 59244376L
astr += 'another'
id(astr)
Out[24]: 59947360L

字符串对象在Python中用PyStringObject表示，扩展定义后如下。

typedef struct {
  Py_ssize_t ob_refcnt;            // 引用计数
  struct _typeobject *ob_type;     // 类型指针
  Py_ssize_t ob_size；             // 字符串的长度，不计算C语言中的结尾NULL
  long ob_shash;                   // 字符串的hash值，没有计算则为-1
  int ob_sstate;                   // 字符串对象的状态： 是否interned等
  char ob_sval[1];                 // 保存字符串的内存，默认先分配1个字符，用来保存额外的末尾NULL值
 
  /* Invariants:
   *     ob_sval contains space for 'ob_size+1' elements.
   *     ob_sval[ob_size] == 0.
   *     ob_shash is the hash of the string or -1 if not computed yet.
   *     ob_sstate != 0 iff the string object is in stringobject.c's
   *       'interned' dictionary; in this case the two references
   *       from 'interned' to this object are *not counted* in ob_refcnt.
   */
} PyStringObject;

ob_type字符串的类型指针，实际指向PyString_Type

ob_size保存的是字符串的实际长度，也是通过len(s)返回的长度值。而字符串实际占用的内存是ob_size + 1，因为C语言中需要额外的NULL作为字符串结束标识符。

ob_sval是实际存储字符串的内存，分配时会请求sizeof(PyStringObject)+size的内存，这样以ob_sval开始的内存长度就是size + 1的长度，正好用来存放以NULL结尾的字符串。

ob_shash是字符串的hash值，当字符串用来比较或者作为key时可以加速查找速度，默认值为-1。

static long
string_hash(PyStringObject *a)
{
    register Py_ssize_t len;
    register unsigned char *p;
    register long x;
#ifdef Py_DEBUG
    assert(_Py_HashSecret_Initialized);
#endif
    if (a->ob_shash != -1)
        return a->ob_shash;
    len = Py_SIZE(a);
    /*
      We make the hash of the empty string be 0, rather than using
      (prefix ^ suffix), since this slightly obfuscates the hash secret
    */
    if (len == 0) {
        a->ob_shash = 0;
        return 0;
    }
    p = (unsigned char *) a->ob_sval;
    x = _Py_HashSecret.prefix;
    x ^= *p << 7;
    while (--len >= 0)
        x = (1000003*x) ^ *p++;
    x ^= Py_SIZE(a);
    x ^= _Py_HashSecret.suffix;
    if (x == -1)
        x = -2;
    a->ob_shash = x;
    return x;

ob_sstate记录字符串对象的状态。字符串可能有三种状态：

//Python/objects/stringobject.h
#define SSTATE_NOT_INTERNED 0           // 字符串没有被interned
#define SSTATE_INTERNED_MORTAL 1        // 字符串被interned，可以被回收
#define SSTATE_INTERNED_IMMORTAL 2      // 字符串永久interned，不会被回收

字符串的interned

字符串对象是不可变对象，因此相同的字面值的变量可以绑定到相同的字符串对象上，这样减少了字符串对象的创建次数。这样的行为称为interned。默认情况下空字符串和单字符字符串会被interned。

// Python/objects/stringobject.c
PyObject *
PyString_FromString(const char *str)
{
    register size_t size;
    register PyStringObject *op;
    assert(str != NULL);
    size = strlen(str);
    if (size > PY_SSIZE_T_MAX - PyStringObject_SIZE) {
        PyErr_SetString(PyExc_OverflowError,
            "string is too long for a Python string");
        return NULL;
    }
    // 如果是空字符串直接返回
    if (size == 0 && (op = nullstring) != NULL) {
#ifdef COUNT_ALLOCS
        null_strings++;
#endif
        Py_INCREF(op);
        return (PyObject *)op;
    }
    // 如果是单字符串，先从characters中查找是否存在
    if (size == 1 && (op = characters[*str & UCHAR_MAX]) != NULL) {
#ifdef COUNT_ALLOCS
        one_strings++;
#endif
        Py_INCREF(op);
        return (PyObject *)op;
    }
    /* Inline PyObject_NewVar */
    op = (PyStringObject *)PyObject_MALLOC(PyStringObject_SIZE + size);
    if (op == NULL)
        return PyErr_NoMemory();
    PyObject_INIT_VAR(op, &PyString_Type, size);
    op->ob_shash = -1;
    op->ob_sstate = SSTATE_NOT_INTERNED;
    Py_MEMCPY(op->ob_sval, str, size+1);
    /* share short strings */
    // 空字符串进行interned
    if (size == 0) {
        PyObject *t = (PyObject *)op;
        PyString_InternInPlace(&t);
        op = (PyStringObject *)t;
        nullstring = op;
        Py_INCREF(op);
    // 单字符串保存在characters中，并且进行interned
    } else if (size == 1) {
        PyObject *t = (PyObject *)op;
        PyString_InternInPlace(&t);
        op = (PyStringObject *)t;
        characters[*str & UCHAR_MAX] = op;
        Py_INCREF(op);
    }
    return (PyObject *) op;
}

另外一些情况下，例如__dict__、模块名字等预计会被大量重复使用或者永久使用的字符串，在创建时也会调用PyString_InternInPlace进行interned操作。

// python/objects/stringobject.c
PyAPI_FUNC(PyObject *) PyString_FromString(const char *);
 
PyAPI_FUNC(PyObject *) PyString_FromStringAndSize(const char *, Py_ssize_t);
 
// SSTATE_INTERNED_MORTAL, 计数0会被回收
PyObject *
PyString_InternFromString(const char *cp)
{
    PyObject *s = PyString_FromString(cp);
    if (s == NULL)
        return NULL;
    PyString_InternInPlace(&s);
    return s;
}
 
// SSTATE_INTERNED_IMMORTAL, 永远不会被销毁
void
PyString_InternImmortal(PyObject **p)
{
}
void
PyString_InternInPlace(PyObject **p)
{
    register PyStringObject *s = (PyStringObject *)(*p);
 
    PyObject *t;
 
    // 检查值使用在PyStringObject上, 派生类不适用
    if (s == NULL || !PyString_Check(s))
        Py_FatalError("PyString_InternInPlace: strings only please!");
    /* If it's a string subclass, we don't really know what putting it in the interned dict might do. */
 
    // 不是字符串类型, 返回
    if (!PyString_CheckExact(s))
        return;
    // 本身已经intern了(标志位ob_sstate), 不重复进行, 返回
    if (PyString_CHECK_INTERNED(s))
        return;
 
    // 未初始化字典, 初始化之
    if (interned == NULL) {
        // 注意这里
        interned = PyDict_New();
        if (interned == NULL) {
            PyErr_Clear(); /* Don't leave an exception */
            return;
        }
    }
 
    // 在interned字典中已存在, 修改, 返回intern独享
    t = PyDict_GetItem(interned, (PyObject *)s);
    if (t) {
        Py_INCREF(t);
        Py_DECREF(*p);
        *p = t;
        return;
    }
 
    // 在interned字典中不存在, 放进去
    if (PyDict_SetItem(interned, (PyObject *)s, (PyObject *)s)  0) {
        PyErr_Clear();
        return;
    }
 
    /* 加入interned字典(key-value)会导致refcnt+2,
     * 这里面去掉interned的引用，以使其正确回收
     */
    /* The two references in interned are not counted by refcnt.
       The string deallocator will take care of this */
    Py_REFCNT(s) -= 2;
 
    // 修改字符串对象的ob_sstate标志位, SSTATE_INTERNED_MORTAL
    PyString_CHECK_INTERNED(s) = SSTATE_INTERNED_MORTAL;
}
// 其他代码大量进行intern操作
dict_str = PyString_InternFromString("__dict__")
lenstr = PyString_InternFromString("__len__")
s_true = PyString_InternFromString("true")
empty_array = PyString_InternFromString("[]")

字符串对象的回收

当字符串的引用计数为零时会被回收。

static void
string_dealloc(PyObject *op)
{
    switch (PyString_CHECK_INTERNED(op)) {
        case SSTATE_NOT_INTERNED:
            break;
            
        // 从interned字典中去掉，然后再tp_free掉
        case SSTATE_INTERNED_MORTAL:
            /* revive dead object temporarily for DelItem */
            Py_REFCNT(op) = 3;
            if (PyDict_DelItem(interned, op) != 0)
                Py_FatalError(
                    "deletion of interned string failed");
            break;
        // 这种类型的字符串不会被回收，一直有引用
        case SSTATE_INTERNED_IMMORTAL:
            Py_FatalError("Immortal interned string died.");
        default:
            Py_FatalError("Inconsistent interned string state.");
    }
    Py_TYPE(op)->tp_free(op);
}

字符串对象的其他操作

可以通过字符串对象的类的结构中找到对象的操作函数。

// python/objects/stringobject.c
PyTypeObject PyString_Type = {
    PyVarObject_HEAD_INIT(&PyType_Type, 0)
    "str",
    PyStringObject_SIZE,
    sizeof(char),
    string_dealloc,                             /* tp_dealloc */
    (printfunc)string_print,                    /* tp_print */
    0,                                          /* tp_getattr */
    0,                                          /* tp_setattr */
    0,                                          /* tp_compare */
    string_repr,                                /* tp_repr */
    &string_as_number,                          /* tp_as_number */
    &string_as_sequence,                        /* tp_as_sequence */
    &string_as_mapping,                         /* tp_as_mapping */
    (hashfunc)string_hash,                      /* tp_hash */
    0,                                          /* tp_call */
    string_str,                                 /* tp_str */
    PyObject_GenericGetAttr,                    /* tp_getattro */
    0,                                          /* tp_setattro */
    &string_as_buffer,                          /* tp_as_buffer */
    Py_TPFLAGS_DEFAULT | Py_TPFLAGS_CHECKTYPES |
        Py_TPFLAGS_BASETYPE | Py_TPFLAGS_STRING_SUBCLASS |
        Py_TPFLAGS_HAVE_NEWBUFFER,              /* tp_flags */
    string_doc,                                 /* tp_doc */
    0,                                          /* tp_traverse */
    0,                                          /* tp_clear */
    (richcmpfunc)string_richcompare,            /* tp_richcompare */
    0,                                          /* tp_weaklistoffset */
    0,                                          /* tp_iter */
    0,                                          /* tp_iternext */
    string_methods,                             /* tp_methods */
    0,                                          /* tp_members */
    0,                                          /* tp_getset */
    &PyBaseString_Type,                         /* tp_base */
    0,                                          /* tp_dict */
    0,                                          /* tp_descr_get */
    0,                                          /* tp_descr_set */
    0,                                          /* tp_dictoffset */
    0,                                          /* tp_init */
    0,                                          /* tp_alloc */
    string_new,                                 /* tp_new */
    PyObject_Del,                               /* tp_free */
};

tp_base被赋值为PyBaseString_Type，因此字符串对象是basestring的子类。

(完）

Python源码剖析—整数对象PyIntObject

发表于 2016-10-17 | 分类于 python源码剖析 |

整数对象的结构

整数对象是固定大小的Python对象，内部只有一个ob_ival保存实际的整数值。

typedef struct {
    PyObject_HEAD
    long ob_ival;
} PyIntObject;

整数对象的缓存

为了最大限度的减少内存分配和垃圾回收，Python对整数对象设计了缓存。整数对象的缓存由两种类别构成：

小整数对象: 在Python启动时创建，永远不会回收
其他整数对象：创建时分配，回收时先缓存；在最高代的垃圾回收中整体回收

在Python启动时会创建一批默认值为[5, 257)的小整数对象，存储在small_ints中。这些整数对象的生命周期为Python的生命周期，不会被回收。Python只所以这样处理是因此解释器内部会频繁用到这些小整数，如果每次都分配-回收-再分配显然效率不高，不如创建后一直保留用空间换时间。

#ifndef NSMALLPOSINTS
#define NSMALLPOSINTS           257
#endif
#ifndef NSMALLNEGINTS
#define NSMALLNEGINTS           5
#endif
#if NSMALLNEGINTS + NSMALLPOSINTS > 0
/* References to small integers are saved in this array so that they
   can be shared.
   The integers that are saved are those in the range
   -NSMALLNEGINTS (inclusive) to NSMALLPOSINTS (not inclusive).
*/
static PyIntObject *small_ints[NSMALLNEGINTS + NSMALLPOSINTS];

可以通过id命令查看小整数对象的特性。

id(246)
id(256)
Out[55]: 31363504L
id(256)
Out[56]: 31363504L  # id不会改变
id(257)
Out[53]: 60259096L
id(257)
Out[54]: 60259024L # id会改变

通过上面的例子我们可以知道，其他整数对象使用的内存是不固定的，申请时分配释放时回收。当然，这个回收并不一定是返还给系统内存，整数对象系统本身会缓存一部分整数对象。下面通过整数对象系统的初始化揭露整数的缓存方案。

整数对象的初始化

当Python初始化时会调用_PyInt_Init函数进行整数的初始化。

int
_PyInt_Init(void)
{
    PyIntObject *v;
    int ival;
#if NSMALLNEGINTS + NSMALLPOSINTS > 0
    for (ival = -NSMALLNEGINTS; ival < NSMALLPOSINTS; ival++) {
          if (!free_list && (free_list = fill_free_list()) == NULL)
                    return 0;
        /* PyObject_New is inlined */
        v = free_list;
        free_list = (PyIntObject *)Py_TYPE(v);
        PyObject_INIT(v, &PyInt_Type);
        v->ob_ival = ival;
        small_ints[ival + NSMALLNEGINTS] = v;
    }
#endif
    return 1;
}

缓存会用到数据结构PyIntBlock以及block_list和free_list链表。PyInBlock用来一次申请多个整数对象的内存，然后再一个个用作PyIntObject，并且通过域next链接到block_list链表上。free_list中是空闲的PyIntObject的链表。fill_free_list初始化后的内存结构如下。

然后通过_PyInt_init初始化为小整数，并将其指针存储到samll_ints数组中加快查找。_PyInt_init初始化后的内存结构如下。

我们可以看到整数对象通过PyIntBlock和free_list进行内存申请和缓存的。

整数对象的创建

当新创建一个整数对象时，先从free_list中查找空闲的整数对象，如果有则直接使用；否则会重新分配PyIntBlock结构并进行初始化。

PyObject *
PyInt_FromLong(long ival)
{
    register PyIntObject *v;
#if NSMALLNEGINTS + NSMALLPOSINTS > 0
    if (-NSMALLNEGINTS <= ival && ival < NSMALLPOSINTS) {       # 是小整数则直接使用
        v = small_ints[ival + NSMALLNEGINTS];
        Py_INCREF(v);
#ifdef COUNT_ALLOCS
        if (ival >= 0)
            quick_int_allocs++;
        else
            quick_neg_int_allocs++;
#endif
        return (PyObject *) v;
    }
#endif
    if (free_list == NULL) {
        if ((free_list = fill_free_list()) == NULL)             # 没有空闲的整数对象则分配
            return NULL;
    }
    /* Inline PyObject_New */
    v = free_list;
    free_list = (PyIntObject *)Py_TYPE(v);
    PyObject_INIT(v, &PyInt_Type);
    v->ob_ival = ival;
    return (PyObject *) v;
}

创建一个新的整数257之后的数据结构：

整数对象的回收

当整数对象的引用计数归零时则对其进行回收，由函数int_free操作

static void
int_free(PyIntObject *v)
{
    Py_TYPE(v) = (struct _typeobject *)free_list;
    free_list = v;
}

可以看到被回收的整数对象被连接到free_list链表中。这里有个问题，整数对象的内存什么时候才真正释放呢？

整数对象的释放

原来整数对象的真正释放是在最高代的GC中进行，当GC运行时会调用PyInt_ClearFreeList进行整数对象内存的释放PyInt_ClearFreeList对整个block_list进行遍历，如果其中所有的整数对象的引用计数都为零，则释放整个block。可见整数对象的内存是以PyIntBlock为单位申请和释放的。

int
PyInt_ClearFreeList(void)
{
    PyIntObject *p;
    PyIntBlock *list, *next;
    int i;
    int u;                      /* remaining unfreed ints per block */
    int freelist_size = 0;
    list = block_list;
    block_list = NULL;
    free_list = NULL;
    while (list != NULL) {
        u = 0;
        for (i = 0, p = &list->objects[0];
             i < N_INTOBJECTS;
             i++, p++) {
            if (PyInt_CheckExact(p) && p->ob_refcnt != 0)
                u++;
        }
        next = list->next;
        if (u) {                        # 遍历block发现其有引用计数部位零的对象
            list->next = block_list;
            block_list = list;
            # 将PyIntBlock中引用计数为零的整数对象重新挂到free_list链表中
            for (i = 0, p = &list->objects[0];
                 i < N_INTOBJECTS;
                 i++, p++) {
                if (!PyInt_CheckExact(p) ||
                    p->ob_refcnt == 0) {
                    Py_TYPE(p) = (struct _typeobject *)
                        free_list;
                    free_list = p;
                }
#if NSMALLNEGINTS + NSMALLPOSINTS > 0
               /* 这段代码没有作用。小整数对象都在small_ints中？
                */
                else if (-NSMALLNEGINTS <= p->ob_ival &&
                         p->ob_ival < NSMALLPOSINTS &&
                         small_ints[p->ob_ival +
                                    NSMALLNEGINTS] == NULL) {
                    Py_INCREF(p);
                    small_ints[p->ob_ival +
                               NSMALLNEGINTS] = p;
                }
#endif
            }
        }
        else {              # 整个block的整数对象引用计数均为零，释放整个block
            PyMem_FREE(list);
        }
        freelist_size += u;
        list = next;
    }
    return freelist_size;
}

整数对象的操作符

整数对象定义了许多操作符，可以通过以下代码自行查看。


static PyNumberMethods int_as_number = {
    (binaryfunc)int_add,        /*nb_add*/
    (binaryfunc)int_sub,        /*nb_subtract*/
    (binaryfunc)int_mul,        /*nb_multiply*/
    (binaryfunc)int_classic_div, /*nb_divide*/
    (binaryfunc)int_mod,        /*nb_remainder*/
    (binaryfunc)int_divmod,     /*nb_divmod*/
    (ternaryfunc)int_pow,       /*nb_power*/
    (unaryfunc)int_neg,         /*nb_negative*/
    (unaryfunc)int_int,         /*nb_positive*/
    (unaryfunc)int_abs,         /*nb_absolute*/
    (inquiry)int_nonzero,       /*nb_nonzero*/
    (unaryfunc)int_invert,      /*nb_invert*/
    (binaryfunc)int_lshift,     /*nb_lshift*/
    (binaryfunc)int_rshift,     /*nb_rshift*/
    (binaryfunc)int_and,        /*nb_and*/
    (binaryfunc)int_xor,        /*nb_xor*/
    (binaryfunc)int_or,         /*nb_or*/
    int_coerce,                 /*nb_coerce*/
    (unaryfunc)int_int,         /*nb_int*/
    (unaryfunc)int_long,        /*nb_long*/
    (unaryfunc)int_float,       /*nb_float*/
    (unaryfunc)int_oct,         /*nb_oct*/
    (unaryfunc)int_hex,         /*nb_hex*/
    0,                          /*nb_inplace_add*/
    0,                          /*nb_inplace_subtract*/
    0,                          /*nb_inplace_multiply*/
    0,                          /*nb_inplace_divide*/
    0,                          /*nb_inplace_remainder*/
    0,                          /*nb_inplace_power*/
    0,                          /*nb_inplace_lshift*/
    0,                          /*nb_inplace_rshift*/
    0,                          /*nb_inplace_and*/
    0,                          /*nb_inplace_xor*/
    0,                          /*nb_inplace_or*/
    (binaryfunc)int_div,        /* nb_floor_divide */
    (binaryfunc)int_true_divide, /* nb_true_divide */
    0,                          /* nb_inplace_floor_divide */
    0,                          /* nb_inplace_true_divide */
    (unaryfunc)int_int,         /* nb_index */
};
PyTypeObject PyInt_Type = {
    PyVarObject_HEAD_INIT(&PyType_Type, 0)
    "int",
    sizeof(PyIntObject),
    0,
    (destructor)int_dealloc,                    /* tp_dealloc */
    (printfunc)int_print,                       /* tp_print */
    0,                                          /* tp_getattr */
    0,                                          /* tp_setattr */
    (cmpfunc)int_compare,                       /* tp_compare */
    (reprfunc)int_to_decimal_string,            /* tp_repr */
    &int_as_number,                             /* tp_as_number */
    0,                                          /* tp_as_sequence */
    0,                                          /* tp_as_mapping */
    (hashfunc)int_hash,                         /* tp_hash */
    0,                                          /* tp_call */
    (reprfunc)int_to_decimal_string,            /* tp_str */
    PyObject_GenericGetAttr,                    /* tp_getattro */
    0,                                          /* tp_setattro */
    0,                                          /* tp_as_buffer */
    Py_TPFLAGS_DEFAULT | Py_TPFLAGS_CHECKTYPES |
        Py_TPFLAGS_BASETYPE | Py_TPFLAGS_INT_SUBCLASS,          /* tp_flags */
    int_doc,                                    /* tp_doc */
    0,                                          /* tp_traverse */
    0,                                          /* tp_clear */
    0,                                          /* tp_richcompare */
    0,                                          /* tp_weaklistoffset */
    0,                                          /* tp_iter */
    0,                                          /* tp_iternext */
    int_methods,                                /* tp_methods */
    0,                                          /* tp_members */
    int_getset,                                 /* tp_getset */
    0,                                          /* tp_base */
    0,                                          /* tp_dict */
    0,                                          /* tp_descr_get */
    0,                                          /* tp_descr_set */
    0,                                          /* tp_dictoffset */
    0,                                          /* tp_init */
    0,                                          /* tp_alloc */
    int_new,                                    /* tp_new */
    (freefunc)int_free,                         /* tp_free */
};

intobject.c源码注释

# 略

Python源码剖析—循环垃圾回收器

发表于 2016-10-11 | 分类于 python源码剖析 |

Python垃圾回收概述

Python中的垃圾回收机制基于引用计数(ob_refcnt)，因此需要解决循环引用导致引用计数不能归零的问题。例如

# create[1]
list1=[]              # del
list2=[list1]         # del
list1.append(list2)
list3 = []            # del
list4 = []
list5 = [list3, list4] # del
list6 = [list3, list4]
list3.append(list4)
# del[2]
del list1; del list2;del list3;del list5 # [2]
# list1.ob_refcnt == list2.ob_refcnt == 1
# collect[3]
gc.collect()

虽然list1与list2已经成为需要回收的垃圾，但是由于相互引用导致引用计数不能归零，从而不能触发自动回收。因此Python引入了循环垃圾收集器。

循环垃圾收集器的原理

判断对象是否为垃圾的逻辑比较直白，有外部引用或者被有外部引用的对象引用的对象为非垃圾对象；否则为垃圾对象。具体过程为，遍历所有对象将对象中引用的元素(其他对象)的引用计数减一，最后引用计数不归零的对象(存在外部引用)不是垃圾对象；被不是垃圾对象引用的元素(其他对象)也不是垃圾对象；剩余的则为垃圾对象。可以归纳为如下步骤：

创建可能存在循环应用的对象时，将该对象纳入链表进行管理
遍历所有纳入管理的对象，将对象引用的元素(其他对象)的引用计数减一
再次遍历:
<1> 处理对象：对该对象进行标记
所有引用计数为零的对象没有外部引用，标记为可能是垃圾；
所有引用计数不为零的对象存在外部引用，必然不是垃圾。

|0 |可能是垃圾 |list1、list2、list3、list5|
|>0 |不是垃圾 |list4、list6|

<2> 处理对象：遍历不是垃圾对象中的元素，不是垃圾对象中的元素必然不是垃圾
最后没有被确定不是垃圾的对象就是垃圾对象

这部分处理代码比较复杂，每个对象可能作为两种角色进行处理。作为代中的对象以及作为对象中的引用元素。如果作为元素被处理，则肯定不是垃圾。

各个阶段对象中的引用计数

垃圾对象不一定能被自动回收

垃圾对象不一定能被自动回收。所以上面的步骤只能确定垃圾对象，然后对垃圾对象进行额外处理甄别不能回收和能被回收的部分。

垃圾对象：没有外部引用的对象，也没有被有外部引用的对象引用
可回收对象：垃圾对象中能够被自动回收的对象
reachable：非垃圾对象，存在外部引用或者被外部引用的对象引用
unrechable: 垃圾对象
collectable: 可回收对象
finalizers: 垃圾对象中不能被自动回收的对象。一些对象存在析构函数并且相互引用，这样的对象Python不能自动确定回收顺序，因此不能被自动回收。

不是所有对象都纳入循环垃圾收集器

一些基本对象不会产生循环引用，例如int、float、string等，所以没有必须使用循环垃圾收集器，基本的引用计数回收机制即可。还有一些容器类对象，他们中的元素都是基本元素不会引起循环引用，例如{‘a’:1}、(1, 2, 3)，因此也不纳入循环垃圾收集器。所以只有部分容器类对象、生成器、含__del__类等才纳入循环垃圾收集器。

垃圾回收中的代

如上分析，整个循环垃圾收集的效率严重依赖可能引起循环引用的对象的个数。为了减少垃圾回收的动作，Python将对象分代：存活越长的对象越不可能是垃圾，就减少对其进行垃圾回收的次数。那么存活的时间长短就用经过了几次垃圾回收来判断，于是刚创建的对象为一代，当经过一次垃圾回收还存活的对象放入二代；多次一代垃圾回收后，才进行一次二代垃圾回收。Python将整个对象分为三代，当分配足够数量的对象后(700)进行一次一代回收；当进行一定数量(10)一代回收后进行二代回收；同理进行三代回收。

gcmodule.c源码分析

/*
  Reference Cycle Garbage Collection
  ==================================
  Neil Schemenauer <nas@arctrix.com>
  Based on a post on the python-dev list.  Ideas from Guido van Rossum,
  Eric Tiedemann, and various others.
  http://www.arctrix.com/nas/python/gc/
  http://www.python.org/pipermail/python-dev/2000-March/003869.html
  http://www.python.org/pipermail/python-dev/2000-March/004010.html
  http://www.python.org/pipermail/python-dev/2000-March/004022.html
  For a highlevel view of the collection process, read the collect
  function.
*/
#include "Python.h"
#include "frameobject.h"        /* for PyFrame_ClearFreeList */
/* 
 * 带有垃圾回收头部的对象的内存模型
 * ------------------------------
 * g          o
 * |PyGC_Head | PyObject | other|
 * ------------------------------
 *           
 * 因此给定地址o计算g就是将地址减少sizeof(PyGC_HEAD)
 * 
 * 说明：
 * 1. 垃圾管理有两部分 引用计数 和 循环垃圾收集器
 * 2. 简单的对象不产生循环引用使用引用计数即可：例如 int、float、string等
 * 3. 有可能产生循环引用的对象(容器类)才需要循环垃圾收集器: 例如 list、dict
 * 4. Python对一些容器类对象进行优化，如果容器内的元素不会产生循环引用，
 *    则不会纳入循环引用管理: 例如 {'a': 1}, [1, 2, 'a']
 */
/* 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))
/*** Global GC state ***/
/* 每一代的寿命是比该代年轻的那代运行的次数
 * 例如 第1代没运行一次，第2代年龄增加一岁
 */
struct gc_generation {
    PyGC_Head head;
    /* 寿命 */
    int threshold; /* collection threshold */
    /* 年龄，到了寿命就该回收了 */
    int count; /* count of allocations or collections of younger
                  generations */
};
/* Python只有3代 */
#define NUM_GENERATIONS 3
#define GEN_HEAD(n) (&generations[n].head)
/* linked lists of container objects */
/* linked lists结构，初始化时每个链表指向自己的头
 * 内存模型如下:
 * ---------------------------
 * g0                        |
 * |next(g0)|prev(g0)|0|700|0|
 * g1                        |
 * |next(g1)|prev(g1)|0|10 |0| 
 * g2                         |
 * |next(g2)|prev(g2)|0|10 |0|
 * ---------------------------
 * generation[i].head.next = generation[i].head.prev = &generation[i]
 * &generation[i] == &generation[i].head
 */
static 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},
};
PyGC_Head *_PyGC_generation0 = GEN_HEAD(0);
static int enabled = 1; /* automatic collection enabled? */
/* true if we are currently running the collector */
static int collecting = 0;
/* list of uncollectable objects  */
/* unreachable:
 * 没有外部引用，只有被彼此循环引用的对象。例如
 * list1=[];list2=[list1];list1.append(list2)
 * del list1; del list2
 * 那么list1和list2不能通过其他对象到达
 *
 * uncollectable:
 * unreachable并且有 __del__ 方法的对象
 *  __del__由用户自定义，可能引用了其他的对象。
 */
static PyObject *garbage = NULL;
/* Python string to use if unhandled exception occurs */
static PyObject *gc_str = NULL;
/* Python string used to look for __del__ attribute. */
static PyObject *delstr = NULL;
/* This is the number of objects who survived the last full collection. It
   approximates the number of long lived objects tracked by the GC.
   (by "full collection", we mean a collection of the oldest generation).
*/
/* 经过一次第三代收集还存活的则为 long_lived 
 * 运行一次第三代的垃圾回收叫 full collection。
 */
static Py_ssize_t long_lived_total = 0;
/* This is the number of objects who survived all "non-full" collections,
   and are awaiting to undergo a full collection for the first time.
*/
/* 第二代回收时，不能被回收而移动到第三代的对象个数 */
static Py_ssize_t long_lived_pending = 0;
/*
   NOTE: about the counting of long-lived objects.
   To limit the cost of garbage collection, there are two strategies;
     - make each collection faster, e.g. by scanning fewer objects
     - do less collections
   This heuristic is about the latter strategy.
   In addition to the various configurable thresholds, we only trigger a
   full collection if the ratio
    long_lived_pending / long_lived_total
   is above a given value (hardwired to 25%).
   The reason is that, while "non-full" collections (i.e., collections of
   the young and middle generations) will always examine roughly the same
   number of objects -- determined by the aforementioned thresholds --,
   the cost of a full collection is proportional to the total number of
   long-lived objects, which is virtually unbounded.
   Indeed, it has been remarked that doing a full collection every
   <constant number> of object creations entails a dramatic performance
   degradation in workloads which consist in creating and storing lots of
   long-lived objects (e.g. building a large list of GC-tracked objects would
   show quadratic performance, instead of linear as expected: see issue #4074).
   Using the above ratio, instead, yields amortized linear performance in
   the total number of objects (the effect of which can be summarized
   thusly: "each full garbage collection is more and more costly as the
   number of objects grows, but we do fewer and fewer of them").
   This heuristic was suggested by Martin von Löwis on python-dev in
   June 2008. His original analysis and proposal can be found at:
    http://mail.python.org/pipermail/python-dev/2008-June/080579.html
*/
/*
   NOTE: about untracking of mutable objects.
   Certain types of container cannot participate in a reference cycle, and
   so do not need to be tracked by the garbage collector. Untracking these
   objects reduces the cost of garbage collections. However, determining
   which objects may be untracked is not free, and the costs must be
   weighed against the benefits for garbage collection.
   There are two possible strategies for when to untrack a container:
   i) When the container is created.
   ii) When the container is examined by the garbage collector.
   Tuples containing only immutable objects (integers, strings etc, and
   recursively, tuples of immutable objects) do not need to be tracked.
   The interpreter creates a large number of tuples, many of which will
   not survive until garbage collection. It is therefore not worthwhile
   to untrack eligible tuples at creation time.
   Instead, all tuples except the empty tuple are tracked when created.
   During garbage collection it is determined whether any surviving tuples
   can be untracked. A tuple can be untracked if all of its contents are
   already not tracked. Tuples are examined for untracking in all garbage
   collection cycles. It may take more than one cycle to untrack a tuple.
   Dictionaries containing only immutable objects also do not need to be
   tracked. Dictionaries are untracked when created. If a tracked item is
   inserted into a dictionary (either as a key or value), the dictionary
   becomes tracked. During a full garbage collection (all generations),
   the collector will untrack any dictionaries whose contents are not
   tracked.
   The module provides the python function is_tracked(obj), which returns
   the CURRENT tracking status of the object. Subsequent garbage
   collections may change the tracking status of the object.
   Untracking of certain containers was introduced in issue #4688, and
   the algorithm was refined in response to issue #14775.
*/
 /* 简单说：
  * 垃圾回收的效率取决于纳入垃圾管理的对象的数量
  */
/* set for debugging information */
#define DEBUG_STATS             (1<<0) /* print collection statistics */
#define DEBUG_COLLECTABLE       (1<<1) /* print collectable objects */
#define DEBUG_UNCOLLECTABLE     (1<<2) /* print uncollectable objects */
#define DEBUG_INSTANCES         (1<<3) /* print instances */
#define DEBUG_OBJECTS           (1<<4) /* print other objects */
#define DEBUG_SAVEALL           (1<<5) /* save all garbage in gc.garbage */
#define DEBUG_LEAK              DEBUG_COLLECTABLE | \
                DEBUG_UNCOLLECTABLE | \
                DEBUG_INSTANCES | \
                DEBUG_OBJECTS | \
                DEBUG_SAVEALL
static int debug;
static PyObject *tmod = NULL;
/*--------------------------------------------------------------------------
gc_refs values.
Between collections, every gc'ed object has one of two gc_refs values:
GC_UNTRACKED
    The initial state; objects returned by PyObject_GC_Malloc are in this
    state.  The object doesn't live in any generation list, and its
    tp_traverse slot must not be called.
    
    没有纳入收集器管理的对象，例如 刚通过PyObject_GC_Malloc初始化的对象，
    int，float等，当然也不存在任何"代"中。
GC_REACHABLE
    The object lives in some generation list, and its tp_traverse is safe to
    call.  An object transitions to GC_REACHABLE when PyObject_GC_Track
    is called.
    纳入了收集器管理的对象，并且可以通过各对象的tp_traverse处理的对象
    纳入管理，并且在有向图中可以到达(不存在循环引用)
代中对象的处理和gc_refs：
1. 遍历整个代中的对象，将对象中的所有元素gc_refs-1
2. 再次遍历整个代中的对象，此时gc_refs：
   0：对象只被代中的其他对象引用，例如：
      list1=[]; list2=[list1];del list1;del list2
      设置gc_refs = GC_TENTATIVELY_UNREACHABLE
   >0: 对象有外部引用
      设置gc_refs = 1
3. 遍历代中的对象中的元素，此时gc_refs：
   GC_TENTATIVELY_UNREACHABLE: 
       说明之前作为代中的对象处理过，确实是unreachable
       设置GC_REACHABLE
   1，GC_UNTRACKED，GC_REACHABLE： 
       保持不变，仍在代中
During a collection, gc_refs can temporarily take on other states:
>= 0
    At the start of a collection, update_refs() copies the true refcount
    to gc_refs, for each object in the generation being collected.
    subtract_refs() then adjusts gc_refs so that it equals the number of
    times an object is referenced directly from outside the generation
    being collected.
    gc_refs remains >= 0 throughout these steps.
GC_TENTATIVELY_UNREACHABLE
    move_unreachable() then moves objects not reachable (whether directly or
    indirectly) from outside the generation into an "unreachable" set.
    Objects that are found to be reachable have gc_refs set to GC_REACHABLE
    again.  Objects that are found to be unreachable have gc_refs set to
    GC_TENTATIVELY_UNREACHABLE.  It's "tentatively" because the pass doing
    this can't be sure until it ends, and GC_TENTATIVELY_UNREACHABLE may
    transition back to GC_REACHABLE.
    遍历时的临时状态，当遍历时将暂时不能到达的对象设置为该状态。
    Only objects with GC_TENTATIVELY_UNREACHABLE still set are candidates
    for collection.  If it's decided not to collect such an object (e.g.,
    it has a __del__ method), its gc_refs is restored to GC_REACHABLE again.
----------------------------------------------------------------------------
*/
#define GC_UNTRACKED                    _PyGC_REFS_UNTRACKED
#define GC_REACHABLE                    _PyGC_REFS_REACHABLE
#define GC_TENTATIVELY_UNREACHABLE      _PyGC_REFS_TENTATIVELY_UNREACHABLE
#define IS_TRACKED(o) ((AS_GC(o))->gc.gc_refs != GC_UNTRACKED)
#define IS_REACHABLE(o) ((AS_GC(o))->gc.gc_refs == GC_REACHABLE)
#define IS_TENTATIVELY_UNREACHABLE(o) ( \
    (AS_GC(o))->gc.gc_refs == GC_TENTATIVELY_UNREACHABLE)
/*** list functions ***/
static void
gc_list_init(PyGC_Head *list)
{
    list->gc.gc_prev = list;
    list->gc.gc_next = list;
}
static int
gc_list_is_empty(PyGC_Head *list)
{
    return (list->gc.gc_next == list);
}
#if 0
/* This became unused after gc_list_move() was introduced. */
/* Append `node` to `list`. */
static void
gc_list_append(PyGC_Head *node, PyGC_Head *list)
{
    node->gc.gc_next = list;
    node->gc.gc_prev = list->gc.gc_prev;
    node->gc.gc_prev->gc.gc_next = node;
    list->gc.gc_prev = node;
}
#endif
/* Remove `node` from the gc list it's currently in. */
static void
gc_list_remove(PyGC_Head *node)
{
    node->gc.gc_prev->gc.gc_next = node->gc.gc_next;
    node->gc.gc_next->gc.gc_prev = node->gc.gc_prev;
    node->gc.gc_next = NULL; /* object is not currently tracked */
}
/* Move `node` from the gc list it's currently in (which is not explicitly
 * named here) to the end of `list`.  This is semantically the same as
 * gc_list_remove(node) followed by gc_list_append(node, list).
 */
static void
gc_list_move(PyGC_Head *node, PyGC_Head *list)
{
    PyGC_Head *new_prev;
    PyGC_Head *current_prev = node->gc.gc_prev;
    PyGC_Head *current_next = node->gc.gc_next;
    /* Unlink from current list. */
    current_prev->gc.gc_next = current_next;
    current_next->gc.gc_prev = current_prev;
    /* Relink at end of new list. */
    new_prev = node->gc.gc_prev = list->gc.gc_prev;
    new_prev->gc.gc_next = list->gc.gc_prev = node;
    node->gc.gc_next = list;
}
/* append list `from` onto list `to`; `from` becomes an empty list */
/* 将 from 整体挂到 to 的头部 */
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 Py_ssize_t
gc_list_size(PyGC_Head *list)
{
    PyGC_Head *gc;
    Py_ssize_t n = 0;
    for (gc = list->gc.gc_next; gc != list; gc = gc->gc.gc_next) {
        n++;
    }
    return n;
}
/* Append objects in a GC list to a Python list.
 * Return 0 if all OK, < 0 if error (out of memory for list).
 */
static int
append_objects(PyObject *py_list, PyGC_Head *gc_list)
{
    PyGC_Head *gc;
    for (gc = gc_list->gc.gc_next; gc != gc_list; gc = gc->gc.gc_next) {
        PyObject *op = FROM_GC(gc);
        if (op != py_list) {
            if (PyList_Append(py_list, op)) {
                return -1; /* exception */
            }
        }
    }
    return 0;
}
/*** end of list stuff ***/
/* Set all gc_refs = ob_refcnt.  After this, gc_refs is > 0 for all objects
 * in containers, and is GC_REACHABLE for all tracked gc objects not in
 * containers.
 */
static void
update_refs(PyGC_Head *containers)
{
    PyGC_Head *gc = containers->gc.gc_next;
    for (; gc != containers; gc = gc->gc.gc_next) {
        assert(gc->gc.gc_refs == GC_REACHABLE);
        gc->gc.gc_refs = Py_REFCNT(FROM_GC(gc));
        /* Python's cyclic gc should never see an incoming refcount
         * of 0:  if something decref'ed to 0, it should have been
         * deallocated immediately at that time.
         * Possible cause (if the assert triggers):  a tp_dealloc
         * routine left a gc-aware object tracked during its teardown
         * phase, and did something-- or allowed something to happen --
         * that called back into Python.  gc can trigger then, and may
         * see the still-tracked dying object.  Before this assert
         * was added, such mistakes went on to allow gc to try to
         * delete the object again.  In a debug build, that caused
         * a mysterious segfault, when _Py_ForgetReference tried
         * to remove the object from the doubly-linked list of all
         * objects a second time.  In a release build, an actual
         * double deallocation occurred, which leads to corruption
         * of the allocator's internal bookkeeping pointers.  That's
         * so serious that maybe this should be a release-build
         * check instead of an assert?
         */
        assert(gc->gc.gc_refs != 0);
    }
}
/* A traversal callback for subtract_refs. */
static int
visit_decref(PyObject *op, void *data)
{
    assert(op != NULL);
    if (PyObject_IS_GC(op)) {
        PyGC_Head *gc = AS_GC(op);
        /* We're only interested in gc_refs for objects in the
         * generation being collected, which can be recognized
         * because only they have positive gc_refs.
         */
        assert(gc->gc.gc_refs != 0); /* else refcount was too small */
        if (gc->gc.gc_refs > 0)
            gc->gc.gc_refs--;
    }
    return 0;
}
/* Subtract internal references from gc_refs.  After this, gc_refs is >= 0
 * for all objects in containers, and is GC_REACHABLE for all tracked gc
 * objects not in containers.  The ones with gc_refs > 0 are directly
 * reachable from outside containers, and so can't be collected.
 */
/*
 * 调用容器类的tp_traverse对每个元素的gc_refs-1，
 * 如果gc_refs>0说明元素在该容器外还有引用，所以不能回收
 */
static void
subtract_refs(PyGC_Head *containers)
{
    traverseproc traverse;
    PyGC_Head *gc = containers->gc.gc_next;
    for (; gc != containers; gc=gc->gc.gc_next) {
        traverse = Py_TYPE(FROM_GC(gc))->tp_traverse;
        (void) traverse(FROM_GC(gc),
                       (visitproc)visit_decref,
                       NULL);
    }
}
/* A traversal callback for move_unreachable. */
/* op 是容器类的元素；reachable是代本身 */
static int
visit_reachable(PyObject *op, PyGC_Head *reachable)
{
    if (PyObject_IS_GC(op)) {
        PyGC_Head *gc = AS_GC(op);
        const Py_ssize_t gc_refs = gc->gc.gc_refs;
        if (gc_refs == 0) {
            /* This is in move_unreachable's 'young' list, but
             * the traversal hasn't yet gotten to it.  All
             * we need to do is tell move_unreachable that it's
             * reachable.
             */
            /*
             * 元素没有引用，说明只有元素所在的容器本身引用该元素
             * 举例： list1 = []; list2=[]; list1.append(list2); 
             *        del list2
             * 
             * list1->ob_refcnt == 1, list->gc_refs == 1
             * list2->ob_refcnt == 1, list->gc_refs == 0
             *
             * 所以：
             * list2作为list1的元素进入当前的处理，依然是reachable
             */
            gc->gc.gc_refs = 1;
        }
        else if (gc_refs == GC_TENTATIVELY_UNREACHABLE) {
            /* This had gc_refs = 0 when move_unreachable got
             * to it, but turns out it's reachable after all.
             * Move it back to move_unreachable's 'young' list,
             * and move_unreachable will eventually get to it
             * again.
             */
            /*
             * 元素之前经历过1次move_unreachable处理并且被认为是unreachable
             * 例如：
             * list1=[];list2=[list1]; list1.append(list2)
             * del list1;
             *
             * 第一次，list1作为容器被move_unreachable处理
             * 会被标记为GC_TENTATIVELY_UNREACHABLE，认为其可以被回收
             *
             * 第二次，list1作为list2的元素会进入到该处
             * 此时证明list1确实是list2的元素，是reachable的
             *
             * 这里的reachable为代链表本身，就是将gc移动到了链表尾端
             */
            gc_list_move(gc, reachable);
            gc->gc.gc_refs = 1;
        }
        /* Else there's nothing to do.
         * If gc_refs > 0, it must be in move_unreachable's 'young'
         * list, and move_unreachable will eventually get to it.
         * If gc_refs == GC_REACHABLE, it's either in some other
         * generation so we don't care about it, or move_unreachable
         * already dealt with it.
         * If gc_refs == GC_UNTRACKED, it must be ignored.
         */
        /*
         * gc_refs > 0 : 处理过了证明是reachable
         * GC_REACHABLE：标记为reachable的元素
         * GC_UNTRACKED：不需要循环垃圾收集器处理的元素
         */
         else {
            assert(gc_refs > 0
                   || gc_refs == GC_REACHABLE
                   || gc_refs == GC_UNTRACKED);
         }
    }
    return 0;
}
/* 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 (gc->gc.gc_refs) {
            /* 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.
             */
            /* 例如：
             * list1=[];list2=[list1];list1.append(list2)
             * del list1;
             * 
             * 经过subtract_refs后：
             * list1.gc_refs==0; list2.gc_refs==1
             * 所以此时的list2有外部的引用，所以是reachable
             */
            PyObject *op = FROM_GC(gc);
            traverseproc traverse = Py_TYPE(op)->tp_traverse;
            assert(gc->gc.gc_refs > 0);
            gc->gc.gc_refs = GC_REACHABLE;
            (void) traverse(op,
                            (visitproc)visit_reachable,
                            (void *)young);
            next = gc->gc.gc_next;
            /* 如果对象op中的元素都是untracked的，标记op为untracked */
            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.
             */
            /* 代中的对象引用计数为0，说明没有外部引用，可能是unreachable
             * 如果再作为对象(容器)的元素被visit_reachable处理到
             * 说明该对象只作为reachable对象的元素，所以也是reachable
             */
            next = gc->gc.gc_next;
            gc_list_move(gc, unreachable);
            gc->gc.gc_refs = GC_TENTATIVELY_UNREACHABLE;
        }
        gc = next;
    }
}
/* Return true if object has a finalization method.
 * CAUTION:  An instance of an old-style class has to be checked for a
 *__del__ method, and earlier versions of this used to call PyObject_HasAttr,
 * which in turn could call the class's __getattr__ hook (if any).  That
 * could invoke arbitrary Python code, mutating the object graph in arbitrary
 * ways, and that was the source of some excruciatingly subtle bugs.
 */
/*
 * 有几种自定义析构函数的方式：
 * 1. 用户类定义了 __del__ 方法
 * 2. 用户类通过定义 __getattr__ 间接定义了 __del__方法
 * 3. 生成器定义了 __exit__方法
 */
static int
has_finalizer(PyObject *op)
{
    if (PyInstance_Check(op)) {
        assert(delstr != NULL);
        return _PyInstance_Lookup(op, delstr) != NULL;
    }
    else if (PyType_HasFeature(op->ob_type, Py_TPFLAGS_HEAPTYPE))
        return op->ob_type->tp_del != NULL;
    else if (PyGen_CheckExact(op))
        return PyGen_NeedsFinalizing((PyGenObject *)op);
    else
        return 0;
}
/* Try to untrack all currently tracked dictionaries */
static void
untrack_dicts(PyGC_Head *head)
{
    PyGC_Head *next, *gc = head->gc.gc_next;
    while (gc != head) {
        PyObject *op = FROM_GC(gc);
        next = gc->gc.gc_next;
        if (PyDict_CheckExact(op))
            _PyDict_MaybeUntrack(op);
        gc = next;
    }
}
/* unreachable 链表中对象的状态为GC_TENTATIVELY_UNREACHABLE:
 * 1. 具有 __del__ 方法，移动到 finalizers 链表修改状态为 GC_REACHABLE
 * 2. 没有 __del__ 方法，保持不变
 */
/* Move the objects in unreachable with __del__ methods 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_finalizers(PyGC_Head *unreachable, PyGC_Head *finalizers)
{
    PyGC_Head *gc;
    PyGC_Head *next;
    /* March over unreachable.  Move objects with finalizers into
     * `finalizers`.
     */
    /* 将 unreachable 链表中有析构函数的对象移到 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_finalizer(op)) {
            gc_list_move(gc, finalizers);
            gc->gc.gc_refs = GC_REACHABLE;
        }
    }
}
/* A traversal callback for move_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);
            gc->gc.gc_refs = GC_REACHABLE;
        }
    }
    return 0;
}
/* Move objects that are reachable from finalizers, from the unreachable set
 * into finalizers set.
 */
/*
 * 将 finalizers 中的对象设置为 GC_REACHABLE
 */
static void
move_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);
    }
}
/* Clear all weakrefs to unreachable objects, and if such a weakref has a
 * callback, invoke it if necessary.  Note that it's possible for such
 * weakrefs to be outside the unreachable set -- indeed, those are precisely
 * the weakrefs whose callbacks must be invoked.  See gc_weakref.txt for
 * overview & some details.  Some weakrefs with callbacks may be reclaimed
 * directly by this routine; the number reclaimed is the return value.  Other
 * weakrefs with callbacks may be moved into the `old` generation.  Objects
 * moved into `old` have gc_refs set to GC_REACHABLE; the objects remaining in
 * unreachable are left at GC_TENTATIVELY_UNREACHABLE.  When this returns,
 * no object in `unreachable` is weakly referenced anymore.
 */
static int
handle_weakrefs(PyGC_Head *unreachable, PyGC_Head *old)
{
    PyGC_Head *gc;
    PyObject *op;               /* generally FROM_GC(gc) */
    PyWeakReference *wr;        /* generally a cast of op */
    PyGC_Head wrcb_to_call;     /* weakrefs with callbacks to call */
    PyGC_Head *next;
    int num_freed = 0;
    gc_list_init(&wrcb_to_call);
    /* Clear all weakrefs to the objects in unreachable.  If such a weakref
     * also has a callback, move it into `wrcb_to_call` if the callback
     * needs to be invoked.  Note that we cannot invoke any callbacks until
     * all weakrefs to unreachable objects are cleared, lest the callback
     * resurrect an unreachable object via a still-active weakref.  We
     * make another pass over wrcb_to_call, invoking callbacks, after this
     * pass completes.
     */
    for (gc = unreachable->gc.gc_next; gc != unreachable; gc = next) {
        PyWeakReference **wrlist;
        op = FROM_GC(gc);
        assert(IS_TENTATIVELY_UNREACHABLE(op));
        next = gc->gc.gc_next;
        if (! PyType_SUPPORTS_WEAKREFS(Py_TYPE(op)))
            continue;
        /* It supports weakrefs.  Does it have any? */
        wrlist = (PyWeakReference **)
                                PyObject_GET_WEAKREFS_LISTPTR(op);
        /* `op` may have some weakrefs.  March over the list, clear
         * all the weakrefs, and move the weakrefs with callbacks
         * that must be called into wrcb_to_call.
         */
        for (wr = *wrlist; wr != NULL; wr = *wrlist) {
            PyGC_Head *wrasgc;                  /* AS_GC(wr) */
            /* _PyWeakref_ClearRef clears the weakref but leaves
             * the callback pointer intact.  Obscure:  it also
             * changes *wrlist.
             */
            assert(wr->wr_object == op);
            _PyWeakref_ClearRef(wr);
            assert(wr->wr_object == Py_None);
            if (wr->wr_callback == NULL)
                continue;                       /* no callback */
    /* Headache time.  `op` is going away, and is weakly referenced by
     * `wr`, which has a callback.  Should the callback be invoked?  If wr
     * is also trash, no:
     *
     * 1. There's no need to call it.  The object and the weakref are
     *    both going away, so it's legitimate to pretend the weakref is
     *    going away first.  The user has to ensure a weakref outlives its
     *    referent if they want a guarantee that the wr callback will get
     *    invoked.
     *
     * 2. It may be catastrophic to call it.  If the callback is also in
     *    cyclic trash (CT), then although the CT is unreachable from
     *    outside the current generation, CT may be reachable from the
     *    callback.  Then the callback could resurrect insane objects.
     *
     * Since the callback is never needed and may be unsafe in this case,
     * wr is simply left in the unreachable set.  Note that because we
     * already called _PyWeakref_ClearRef(wr), its callback will never
     * trigger.
     *
     * OTOH, if wr isn't part of CT, we should invoke the callback:  the
     * weakref outlived the trash.  Note that since wr isn't CT in this
     * case, its callback can't be CT either -- wr acted as an external
     * root to this generation, and therefore its callback did too.  So
     * nothing in CT is reachable from the callback either, so it's hard
     * to imagine how calling it later could create a problem for us.  wr
     * is moved to wrcb_to_call in this case.
     */
            if (IS_TENTATIVELY_UNREACHABLE(wr))
                continue;
            assert(IS_REACHABLE(wr));
            /* Create a new reference so that wr can't go away
             * before we can process it again.
             */
            Py_INCREF(wr);
            /* Move wr to wrcb_to_call, for the next pass. */
            wrasgc = AS_GC(wr);
            assert(wrasgc != next); /* wrasgc is reachable, but
                                       next isn't, so they can't
                                       be the same */
            gc_list_move(wrasgc, &wrcb_to_call);
        }
    }
    /* Invoke the callbacks we decided to honor.  It's safe to invoke them
     * because they can't reference unreachable objects.
     */
    while (! gc_list_is_empty(&wrcb_to_call)) {
        PyObject *temp;
        PyObject *callback;
        gc = wrcb_to_call.gc.gc_next;
        op = FROM_GC(gc);
        assert(IS_REACHABLE(op));
        assert(PyWeakref_Check(op));
        wr = (PyWeakReference *)op;
        callback = wr->wr_callback;
        assert(callback != NULL);
        /* copy-paste of weakrefobject.c's handle_callback() */
        temp = PyObject_CallFunctionObjArgs(callback, wr, NULL);
        if (temp == NULL)
            PyErr_WriteUnraisable(callback);
        else
            Py_DECREF(temp);
        /* Give up the reference we created in the first pass.  When
         * op's refcount hits 0 (which it may or may not do right now),
         * op's tp_dealloc will decref op->wr_callback too.  Note
         * that the refcount probably will hit 0 now, and because this
         * weakref was reachable to begin with, gc didn't already
         * add it to its count of freed objects.  Example:  a reachable
         * weak value dict maps some key to this reachable weakref.
         * The callback removes this key->weakref mapping from the
         * dict, leaving no other references to the weakref (excepting
         * ours).
         */
        Py_DECREF(op);
        if (wrcb_to_call.gc.gc_next == gc) {
            /* object is still alive -- move it */
            gc_list_move(gc, old);
        }
        else
            ++num_freed;
    }
    return num_freed;
}
static void
debug_instance(char *msg, PyInstanceObject *inst)
{
    char *cname;
    /* simple version of instance_repr */
    PyObject *classname = inst->in_class->cl_name;
    if (classname != NULL && PyString_Check(classname))
        cname = PyString_AsString(classname);
    else
        cname = "?";
    PySys_WriteStderr("gc: %.100s <%.100s instance at %p>\n",
                      msg, cname, inst);
}
static void
debug_cycle(char *msg, PyObject *op)
{
    if ((debug & DEBUG_INSTANCES) && PyInstance_Check(op)) {
        debug_instance(msg, (PyInstanceObject *)op);
    }
    else if (debug & DEBUG_OBJECTS) {
        PySys_WriteStderr("gc: %.100s <%.100s %p>\n",
                          msg, Py_TYPE(op)->tp_name, op);
    }
}
/* Handle uncollectable garbage (cycles with finalizers, 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.
 */
/*
 * 将有 finalizer 的对象放入到 garbage
 * 剩余的 finalizers 中的所有对象放入下一代
 * 这个时候：
 * 1. 没有有析构函数的放入 garbage
 * 2. 有析构函数的放入下一代
 */
static int
handle_finalizers(PyGC_Head *finalizers, PyGC_Head *old)
{
    PyGC_Head *gc = finalizers->gc.gc_next;
    if (garbage == NULL) {
        garbage = PyList_New(0);
        if (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 ((debug & DEBUG_SAVEALL) || has_finalizer(op)) {
            if (PyList_Append(garbage, op) < 0)
                return -1;
        }
    }
    gc_list_merge(finalizers, old);
    return 0;
}
/* 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.
 */
/* 回收garbage */
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);
        assert(IS_TENTATIVELY_UNREACHABLE(op));
        if (debug & DEBUG_SAVEALL) {
            PyList_Append(garbage, op);
        }
        else {
            if ((clear = Py_TYPE(op)->tp_clear) != NULL) {
                Py_INCREF(op);
                clear(op);
                Py_DECREF(op);
            }
        }
        /* clear 函数没有将自己从collectable链表中摘下来
         * 说明还不能clear，则放入下一代
         */
        if (collectable->gc.gc_next == gc) {
            /* object is still alive, move it, it may die later */
            gc_list_move(gc, old);
            gc->gc.gc_refs = GC_REACHABLE;
        }
    }
}
/* Clear all free lists
 * All free lists are cleared during the collection of the highest generation.
 * Allocated items in the free list may keep a pymalloc arena occupied.
 * Clearing the free lists may give back memory to the OS earlier.
 */
/*
 * 回收没有引用的lists
 */
static void
clear_freelists(void)
{
    (void)PyMethod_ClearFreeList();
    (void)PyFrame_ClearFreeList();
    (void)PyCFunction_ClearFreeList();
    (void)PyTuple_ClearFreeList();
#ifdef Py_USING_UNICODE
    (void)PyUnicode_ClearFreeList();
#endif
    (void)PyInt_ClearFreeList();
    (void)PyFloat_ClearFreeList();
}
static double
get_time(void)
{
    double result = 0;
    if (tmod != NULL) {
        PyObject *f = PyObject_CallMethod(tmod, "time", NULL);
        if (f == NULL) {
            PyErr_Clear();
        }
        else {
            if (PyFloat_Check(f))
                result = PyFloat_AsDouble(f);
            Py_DECREF(f);
        }
    }
    return result;
}
/* This is the main function.  Read this to understand how the
 * collection process works. */
static Py_ssize_t
collect(int generation)
{
    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;
    double t1 = 0.0;
    if (delstr == NULL) {
        delstr = PyString_InternFromString("__del__");
        if (delstr == NULL)
            Py_FatalError("gc couldn't allocate \"__del__\"");
    }
    if (debug & DEBUG_STATS) {
        PySys_WriteStderr("gc: collecting generation %d...\n",
                          generation);
        PySys_WriteStderr("gc: objects in each generation:");
        for (i = 0; i < NUM_GENERATIONS; i++)
            PySys_WriteStderr(" %" PY_FORMAT_SIZE_T "d",
                              gc_list_size(GEN_HEAD(i)));
        t1 = get_time();
        PySys_WriteStderr("\n");
    }
    /* update collection and allocation counters */
    /* 递增老一代的年龄 */
    if (generation+1 < NUM_GENERATIONS)
        generations[generation+1].count += 1;
    /* 年轻的所有代都归零，因为会处理所有年轻代的对象*/
    for (i = 0; i <= generation; i++)
        generations[i].count = 0;
    /* merge younger generations with one we are currently collecting */
    /* 收集第3代，则将第1,2代的对象都加入到第3代中 */
    for (i = 0; i < generation; i++) {
        gc_list_merge(GEN_HEAD(i), GEN_HEAD(generation));
    }
    /* handy references */
    young = GEN_HEAD(generation);
    if (generation < NUM_GENERATIONS-1)
        old = GEN_HEAD(generation+1);
    else
        old = young;
    /* 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).
     */
    /* 将当前代的对象的引用计数复制到gc_refs=ob_refcnt */
    update_refs(young);
    
    /* 将本代的所有容器内的元素gc_refs-1 */
    subtract_refs(young);
    /* 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.
     */
    /* 将unreachable的容器放入unreachable链表中 */
    gc_list_init(&unreachable);
    move_unreachable(young, &unreachable);
    /* Move reachable objects to next generation. */
    /* 将剩余的reachable的容器放入老一代中 
     * 如果当前是第2代，则累计long_lived_pending
     */
    if (young != old) {
        if (generation == NUM_GENERATIONS - 2) {
            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);
        long_lived_pending = 0;
        long_lived_total = gc_list_size(young);
    }
    /* All objects in unreachable are trash, but objects reachable from
     * finalizers can't safely be deleted.  Python programmers should take
     * care not to create such things.  For Python, finalizers means
     * instance objects with __del__ methods.  Weakrefs with callbacks
     * can also call arbitrary Python code but they will be dealt with by
     * handle_weakrefs().
     */
    /*
     * unreachable中有析构函数的不能直接清除，所以需要移动到finalizers
     * 将finalizers中的容器中的元素标记为reachable
     */
    gc_list_init(&finalizers);
    move_finalizers(&unreachable, &finalizers);
    /* finalizers contains the unreachable objects with a finalizer;
     * unreachable objects reachable *from* those are also uncollectable,
     * and we move those into the finalizers list too.
     */
    /*
     * finalizers中的对象中的元素也需要加入到finalizers，例如
     * a = 0; class Test(object): def __del__(self): a; del a
     * 如果Test是unreachable，那么a即使是reachable也不能被收集，
     * 所以需要把其也加入到finalizers
     */
    move_finalizer_reachable(&finalizers);
    /* Collect statistics on collectable objects found and print
     * debugging information.
     */
    for (gc = unreachable.gc.gc_next; gc != &unreachable;
                    gc = gc->gc.gc_next) {
        m++;
        if (debug & DEBUG_COLLECTABLE) {
            debug_cycle("collectable", FROM_GC(gc));
        }
    }
    /* Clear weakrefs and invoke callbacks as necessary. */
    m += handle_weakrefs(&unreachable, old);
    /* 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.
     */
    /* 清除可以清除的；把不能清除的放入老一代 */
    delete_garbage(&unreachable, old);
    /* Collect statistics on uncollectable objects found and print
     * debugging information. */
    for (gc = finalizers.gc.gc_next;
         gc != &finalizers;
         gc = gc->gc.gc_next) {
        n++;
        if (debug & DEBUG_UNCOLLECTABLE)
            debug_cycle("uncollectable", FROM_GC(gc));
    }
    if (debug & DEBUG_STATS) {
        double t2 = get_time();
        if (m == 0 && n == 0)
            PySys_WriteStderr("gc: done");
        else
            PySys_WriteStderr(
                "gc: done, "
                "%" PY_FORMAT_SIZE_T "d unreachable, "
                "%" PY_FORMAT_SIZE_T "d uncollectable",
                n+m, n);
        if (t1 && t2) {
            PySys_WriteStderr(", %.4fs elapsed", t2-t1);
        }
        PySys_WriteStderr(".\n");
    }
    /* 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.
     */
    /*
     * 所以如果析构函数中有循环引用，那么可能永远不可能被清除
     */
    (void)handle_finalizers(&finalizers, old);
    /* Clear free list only during the collection of the highest
     * generation */
    if (generation == NUM_GENERATIONS-1) {
        clear_freelists();
    }
    if (PyErr_Occurred()) {
        if (gc_str == NULL)
            gc_str = PyString_FromString("garbage collection");
        PyErr_WriteUnraisable(gc_str);
        Py_FatalError("unexpected exception during garbage collection");
    }
    return n+m;
}
static Py_ssize_t
collect_generations(void)
{
    int i;
    Py_ssize_t n = 0;
    /* Find the oldest generation (highest numbered) where the count
     * exceeds the threshold.  Objects in the that generation and
     * generations younger than it will be collected. */
    /*
     * 垃圾回收的规则：
     * 1. 每代的寿命到了才启动该代的垃圾回收(count>threshold)
     *    (第一代的寿命作为启动垃圾回收的入口)
     * 2. 每代运行一次，则老一代年龄增长一岁
     * 4. 每代运行时，处理所有比其年轻的代的对象
     */ 
    for (i = NUM_GENERATIONS-1; i >= 0; i--) {
        if (generations[i].count > generations[i].threshold) {
            /* Avoid quadratic performance degradation in number
               of tracked objects. See comments at the beginning
               of this file, and issue #4074.
            */
            /* long_lived_pending:
             * 在运行第三代收集之前，从第二代放入第三代的对象个数
             * long_lived_total:
             * 在运行第三代收集时，第三代中不能回收的对象个数
             *
             * 简单来说：第三代中新增加的对象数量大于25%才运行
             */
            if (i == NUM_GENERATIONS - 1
                && long_lived_pending < long_lived_total / 4)
                continue;
            n = collect(i);
            break;
        }
    }
    return n;
}
PyDoc_STRVAR(gc_enable__doc__,
"enable() -> None\n"
"\n"
"Enable automatic garbage collection.\n");
static PyObject *
gc_enable(PyObject *self, PyObject *noargs)
{
    enabled = 1;
    Py_INCREF(Py_None);
    return Py_None;
}
PyDoc_STRVAR(gc_disable__doc__,
"disable() -> None\n"
"\n"
"Disable automatic garbage collection.\n");
static PyObject *
gc_disable(PyObject *self, PyObject *noargs)
{
    enabled = 0;
    Py_INCREF(Py_None);
    return Py_None;
}
PyDoc_STRVAR(gc_isenabled__doc__,
"isenabled() -> status\n"
"\n"
"Returns true if automatic garbage collection is enabled.\n");
static PyObject *
gc_isenabled(PyObject *self, PyObject *noargs)
{
    return PyBool_FromLong((long)enabled);
}
PyDoc_STRVAR(gc_collect__doc__,
"collect([generation]) -> n\n"
"\n"
"With no arguments, run a full collection.  The optional argument\n"
"may be an integer specifying which generation to collect.  A ValueError\n"
"is raised if the generation number is invalid.\n\n"
"The number of unreachable objects is returned.\n");
static PyObject *
gc_collect(PyObject *self, PyObject *args, PyObject *kws)
{
    static char *keywords[] = {"generation", NULL};
    int genarg = NUM_GENERATIONS - 1;
    Py_ssize_t n;
    if (!PyArg_ParseTupleAndKeywords(args, kws, "|i", keywords, &genarg))
        return NULL;
    else if (genarg < 0 || genarg >= NUM_GENERATIONS) {
        PyErr_SetString(PyExc_ValueError, "invalid generation");
        return NULL;
    }
    if (collecting)
        n = 0; /* already collecting, don't do anything */
    else {
        collecting = 1;
        n = collect(genarg);
        collecting = 0;
    }
    return PyInt_FromSsize_t(n);
}
PyDoc_STRVAR(gc_set_debug__doc__,
"set_debug(flags) -> None\n"
"\n"
"Set the garbage collection debugging flags. Debugging information is\n"
"written to sys.stderr.\n"
"\n"
"flags is an integer and can have the following bits turned on:\n"
"\n"
"  DEBUG_STATS - Print statistics during collection.\n"
"  DEBUG_COLLECTABLE - Print collectable objects found.\n"
"  DEBUG_UNCOLLECTABLE - Print unreachable but uncollectable objects found.\n"
"  DEBUG_INSTANCES - Print instance objects.\n"
"  DEBUG_OBJECTS - Print objects other than instances.\n"
"  DEBUG_SAVEALL - Save objects to gc.garbage rather than freeing them.\n"
"  DEBUG_LEAK - Debug leaking programs (everything but STATS).\n");
static PyObject *
gc_set_debug(PyObject *self, PyObject *args)
{
    if (!PyArg_ParseTuple(args, "i:set_debug", &debug))
        return NULL;
    Py_INCREF(Py_None);
    return Py_None;
}
PyDoc_STRVAR(gc_get_debug__doc__,
"get_debug() -> flags\n"
"\n"
"Get the garbage collection debugging flags.\n");
static PyObject *
gc_get_debug(PyObject *self, PyObject *noargs)
{
    return Py_BuildValue("i", debug);
}
PyDoc_STRVAR(gc_set_thresh__doc__,
"set_threshold(threshold0, [threshold1, threshold2]) -> None\n"
"\n"
"Sets the collection thresholds.  Setting threshold0 to zero disables\n"
"collection.\n");
static PyObject *
gc_set_thresh(PyObject *self, PyObject *args)
{
    int i;
    if (!PyArg_ParseTuple(args, "i|ii:set_threshold",
                          &generations[0].threshold,
                          &generations[1].threshold,
                          &generations[2].threshold))
        return NULL;
    for (i = 2; i < NUM_GENERATIONS; i++) {
        /* generations higher than 2 get the same threshold */
        generations[i].threshold = generations[2].threshold;
    }
    Py_INCREF(Py_None);
    return Py_None;
}
PyDoc_STRVAR(gc_get_thresh__doc__,
"get_threshold() -> (threshold0, threshold1, threshold2)\n"
"\n"
"Return the current collection thresholds\n");
static PyObject *
gc_get_thresh(PyObject *self, PyObject *noargs)
{
    return Py_BuildValue("(iii)",
                         generations[0].threshold,
                         generations[1].threshold,
                         generations[2].threshold);
}
PyDoc_STRVAR(gc_get_count__doc__,
"get_count() -> (count0, count1, count2)\n"
"\n"
"Return the current collection counts\n");
static PyObject *
gc_get_count(PyObject *self, PyObject *noargs)
{
    return Py_BuildValue("(iii)",
                         generations[0].count,
                         generations[1].count,
                         generations[2].count);
}
static int
referrersvisit(PyObject* obj, PyObject *objs)
{
    Py_ssize_t i;
    for (i = 0; i < PyTuple_GET_SIZE(objs); i++)
        if (PyTuple_GET_ITEM(objs, i) == obj)
            return 1;
    return 0;
}
static int
gc_referrers_for(PyObject *objs, PyGC_Head *list, PyObject *resultlist)
{
    PyGC_Head *gc;
    PyObject *obj;
    traverseproc traverse;
    for (gc = list->gc.gc_next; gc != list; gc = gc->gc.gc_next) {
        obj = FROM_GC(gc);
        traverse = Py_TYPE(obj)->tp_traverse;
        if (obj == objs || obj == resultlist)
            continue;
        if (traverse(obj, (visitproc)referrersvisit, objs)) {
            if (PyList_Append(resultlist, obj) < 0)
                return 0; /* error */
        }
    }
    return 1; /* no error */
}
PyDoc_STRVAR(gc_get_referrers__doc__,
"get_referrers(*objs) -> list\n\
Return the list of objects that directly refer to any of objs.");
static PyObject *
gc_get_referrers(PyObject *self, PyObject *args)
{
    int i;
    PyObject *result = PyList_New(0);
    if (!result) return NULL;
    for (i = 0; i < NUM_GENERATIONS; i++) {
        if (!(gc_referrers_for(args, GEN_HEAD(i), result))) {
            Py_DECREF(result);
            return NULL;
        }
    }
    return result;
}
/* Append obj to list; return true if error (out of memory), false if OK. */
static int
referentsvisit(PyObject *obj, PyObject *list)
{
    return PyList_Append(list, obj) < 0;
}
PyDoc_STRVAR(gc_get_referents__doc__,
"get_referents(*objs) -> list\n\
Return the list of objects that are directly referred to by objs.");
static PyObject *
gc_get_referents(PyObject *self, PyObject *args)
{
    Py_ssize_t i;
    PyObject *result = PyList_New(0);
    if (result == NULL)
        return NULL;
    for (i = 0; i < PyTuple_GET_SIZE(args); i++) {
        traverseproc traverse;
        PyObject *obj = PyTuple_GET_ITEM(args, i);
        if (! PyObject_IS_GC(obj))
            continue;
        traverse = Py_TYPE(obj)->tp_traverse;
        if (! traverse)
            continue;
        if (traverse(obj, (visitproc)referentsvisit, result)) {
            Py_DECREF(result);
            return NULL;
        }
    }
    return result;
}
PyDoc_STRVAR(gc_get_objects__doc__,
"get_objects() -> [...]\n"
"\n"
"Return a list of objects tracked by the collector (excluding the list\n"
"returned).\n");
static PyObject *
gc_get_objects(PyObject *self, PyObject *noargs)
{
    int i;
    PyObject* result;
    result = PyList_New(0);
    if (result == NULL)
        return NULL;
    for (i = 0; i < NUM_GENERATIONS; i++) {
        if (append_objects(result, GEN_HEAD(i))) {
            Py_DECREF(result);
            return NULL;
        }
    }
    return result;
}
PyDoc_STRVAR(gc_is_tracked__doc__,
"is_tracked(obj) -> bool\n"
"\n"
"Returns true if the object is tracked by the garbage collector.\n"
"Simple atomic objects will return false.\n"
);
static PyObject *
gc_is_tracked(PyObject *self, PyObject *obj)
{
    PyObject *result;
    if (PyObject_IS_GC(obj) && IS_TRACKED(obj))
        result = Py_True;
    else
        result = Py_False;
    Py_INCREF(result);
    return result;
}
PyDoc_STRVAR(gc__doc__,
"This module provides access to the garbage collector for reference cycles.\n"
"\n"
"enable() -- Enable automatic garbage collection.\n"
"disable() -- Disable automatic garbage collection.\n"
"isenabled() -- Returns true if automatic collection is enabled.\n"
"collect() -- Do a full collection right now.\n"
"get_count() -- Return the current collection counts.\n"
"set_debug() -- Set debugging flags.\n"
"get_debug() -- Get debugging flags.\n"
"set_threshold() -- Set the collection thresholds.\n"
"get_threshold() -- Return the current the collection thresholds.\n"
"get_objects() -- Return a list of all objects tracked by the collector.\n"
"is_tracked() -- Returns true if a given object is tracked.\n"
"get_referrers() -- Return the list of objects that refer to an object.\n"
"get_referents() -- Return the list of objects that an object refers to.\n");
static PyMethodDef GcMethods[] = {
    {"enable",             gc_enable,     METH_NOARGS,  gc_enable__doc__},
    {"disable",            gc_disable,    METH_NOARGS,  gc_disable__doc__},
    {"isenabled",          gc_isenabled,  METH_NOARGS,  gc_isenabled__doc__},
    {"set_debug",          gc_set_debug,  METH_VARARGS, gc_set_debug__doc__},
    {"get_debug",          gc_get_debug,  METH_NOARGS,  gc_get_debug__doc__},
    {"get_count",          gc_get_count,  METH_NOARGS,  gc_get_count__doc__},
    {"set_threshold",  gc_set_thresh, METH_VARARGS, gc_set_thresh__doc__},
    {"get_threshold",  gc_get_thresh, METH_NOARGS,  gc_get_thresh__doc__},
    {"collect",            (PyCFunction)gc_collect,
        METH_VARARGS | METH_KEYWORDS,           gc_collect__doc__},
    {"get_objects",    gc_get_objects,METH_NOARGS,  gc_get_objects__doc__},
    {"is_tracked",     gc_is_tracked, METH_O,       gc_is_tracked__doc__},
    {"get_referrers",  gc_get_referrers, METH_VARARGS,
        gc_get_referrers__doc__},
    {"get_referents",  gc_get_referents, METH_VARARGS,
        gc_get_referents__doc__},
    {NULL,      NULL}           /* Sentinel */
};
PyMODINIT_FUNC
initgc(void)
{
    PyObject *m;
    m = Py_InitModule4("gc",
                          GcMethods,
                          gc__doc__,
                          NULL,
                          PYTHON_API_VERSION);
    if (m == NULL)
        return;
    if (garbage == NULL) {
        garbage = PyList_New(0);
        if (garbage == NULL)
            return;
    }
    Py_INCREF(garbage);
    if (PyModule_AddObject(m, "garbage", garbage) < 0)
        return;
    /* Importing can't be done in collect() because collect()
     * can be called via PyGC_Collect() in Py_Finalize().
     * This wouldn't be a problem, except that <initialized> is
     * reset to 0 before calling collect which trips up
     * the import and triggers an assertion.
     */
    if (tmod == NULL) {
        tmod = PyImport_ImportModuleNoBlock("time");
        if (tmod == NULL)
            PyErr_Clear();
    }
#define ADD_INT(NAME) if (PyModule_AddIntConstant(m, #NAME, NAME) < 0) return
    ADD_INT(DEBUG_STATS);
    ADD_INT(DEBUG_COLLECTABLE);
    ADD_INT(DEBUG_UNCOLLECTABLE);
    ADD_INT(DEBUG_INSTANCES);
    ADD_INT(DEBUG_OBJECTS);
    ADD_INT(DEBUG_SAVEALL);
    ADD_INT(DEBUG_LEAK);
#undef ADD_INT
}
/* API to invoke gc.collect() from C */
Py_ssize_t
PyGC_Collect(void)
{
    Py_ssize_t n;
    if (collecting)
        n = 0; /* already collecting, don't do anything */
    else {
        collecting = 1;
        n = collect(NUM_GENERATIONS - 1);
        collecting = 0;
    }
    return n;
}
/* for debugging */
void
_PyGC_Dump(PyGC_Head *g)
{
    _PyObject_Dump(FROM_GC(g));
}
/* extension modules might be compiled with GC support so these
   functions must always be available */
#undef PyObject_GC_Track
#undef PyObject_GC_UnTrack
#undef PyObject_GC_Del
#undef _PyObject_GC_Malloc
void
PyObject_GC_Track(void *op)
{
    _PyObject_GC_TRACK(op);
}
/* for binary compatibility with 2.2 */
void
_PyObject_GC_Track(PyObject *op)
{
    PyObject_GC_Track(op);
}
void
PyObject_GC_UnTrack(void *op)
{
    /* Obscure:  the Py_TRASHCAN mechanism requires that we be able to
     * call PyObject_GC_UnTrack twice on an object.
     */
    if (IS_TRACKED(op))
        _PyObject_GC_UNTRACK(op);
}
/* for binary compatibility with 2.2 */
void
_PyObject_GC_UnTrack(PyObject *op)
{
    PyObject_GC_UnTrack(op);
}
PyObject *
_PyObject_GC_Malloc(size_t basicsize)
{
    PyObject *op;
    PyGC_Head *g;
    if (basicsize > PY_SSIZE_T_MAX - sizeof(PyGC_Head))
        return PyErr_NoMemory();
    g = (PyGC_Head *)PyObject_MALLOC(
        sizeof(PyGC_Head) + basicsize);
    if (g == NULL)
        return PyErr_NoMemory();
    g->gc.gc_refs = GC_UNTRACKED;
    /* 分配的PyObject > 700 就执行垃圾回收 */
    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;
    }
    op = FROM_GC(g);
    return op;
}
PyObject *
_PyObject_GC_New(PyTypeObject *tp)
{
    PyObject *op = _PyObject_GC_Malloc(_PyObject_SIZE(tp));
    if (op != NULL)
        op = PyObject_INIT(op, tp);
    return op;
}
PyVarObject *
_PyObject_GC_NewVar(PyTypeObject *tp, Py_ssize_t nitems)
{
    const size_t size = _PyObject_VAR_SIZE(tp, nitems);
    PyVarObject *op = (PyVarObject *) _PyObject_GC_Malloc(size);
    if (op != NULL)
        op = PyObject_INIT_VAR(op, tp, nitems);
    return op;
}
PyVarObject *
_PyObject_GC_Resize(PyVarObject *op, Py_ssize_t nitems)
{
    const size_t basicsize = _PyObject_VAR_SIZE(Py_TYPE(op), nitems);
    PyGC_Head *g = AS_GC(op);
    if (basicsize > PY_SSIZE_T_MAX - sizeof(PyGC_Head))
        return (PyVarObject *)PyErr_NoMemory();
    g = (PyGC_Head *)PyObject_REALLOC(g,  sizeof(PyGC_Head) + basicsize);
    if (g == NULL)
        return (PyVarObject *)PyErr_NoMemory();
    op = (PyVarObject *) FROM_GC(g);
    Py_SIZE(op) = nitems;
    return op;
}
void
PyObject_GC_Del(void *op)
{
    PyGC_Head *g = AS_GC(op);
    if (IS_TRACKED(op))
        gc_list_remove(g);
    if (generations[0].count > 0) {
        generations[0].count--;
    }
    PyObject_FREE(g);
}
/* for binary compatibility with 2.2 */
#undef _PyObject_GC_Del
void
_PyObject_GC_Del(PyObject *op)
{
    PyObject_GC_Del(op);
}

Python源码剖析—统一内存管理

发表于 2016-10-09 | 分类于 python源码剖析 |

(图片来自: https://nodefe.com/implement-of-pymalloc-from-source/)

arena、pool和block

Python的对象分配器将内存分为三个维度，从大到小叫做arena、pool以及blcok。

arena

一个arena分为两个部分。管理部分arena_object，每次需要创建一个arena时，先创建一个arena_object结构放入arenas数组。然后再申请256KB内存作为arena管理的内存部分。arena_object和arena的内存是分开的，通过域address标记。

pool

将arena的内存按照4KB再划分则为一个个pool。每个pool也分为两部分，内存的高端为pool_header用于管理分配出去的block、回收的block以及从来没有被分配出去的block；剩余的内存作为另一部分再被分为一个个block。每个pool一旦使用只能分配固定个数的block。pool的两部分在同一个连续的页内。

pool会有三种状态：

used: 部分block被分配出去，另一部分还未被分配出去。该状态的pool会被放入usedpools中以加快搜寻可用pool的速度。如果used的pool中的最后的block也被分配出去则pool进入full状态，并且从usedpool中去掉。如果used的pool中的block全被回收则pool进入empty状态，并且从usedpool中去掉放入arena中的freepools链表中。
empy: 所有的block都没有被分配出去。有两种可能，一种是pool中的block都被回收了，从used状态转变而来，这样的pool放入arena的freepools链表中；另外一种是随着arena初始化而来，此时还没有作为pool存在，只是作为arena中没有被使用的内存部分。
full: 所有的block被分配出去了。不存在任何链表中，当有block被回收时进入used状态再放入usedpool中。

block

block是内存管理的最小单位，每次分配需要按照block对齐。每次分配和回收都是固定个数的block。当内存被回收时，所有的内存会放入pool中的链表freeblocks中。没有被分配出去的block存在两个地方，一部分从来没有被分配出去过，通过nextofset表明空闲的block的起始地址；另一部分是分配出去又被回收，会被放入freeblocks中。

被回收的block会将头部作为指针链接下一个被回收的block

1 2	(block )ob = freeblocks freeblocks = &ob

pool的种类

按照每次可以分配的block的个数，pool被分为几种类型(block size)，同时也是其在usedpool中的序号(szidx)。每页为4KB，每个8个block1算作一组，所以pool最多有64个类型。具体可以参见下面代码注释。

python的obmalloc.c源码注释

Objects/obmalloc.c

#include "Python.h"
#if defined(__has_feature)  /* Clang */
 #if __has_feature(address_sanitizer)  /* is ASAN enabled? */
  #define ATTRIBUTE_NO_ADDRESS_SAFETY_ANALYSIS \
        __attribute__((no_address_safety_analysis)) \
        __attribute__ ((noinline))
 #else
  #define ATTRIBUTE_NO_ADDRESS_SAFETY_ANALYSIS
 #endif
#else
 #if defined(__SANITIZE_ADDRESS__)  /* GCC 4.8.x, is ASAN enabled? */
  #define ATTRIBUTE_NO_ADDRESS_SAFETY_ANALYSIS \
        __attribute__((no_address_safety_analysis)) \
        __attribute__ ((noinline))
 #else
  #define ATTRIBUTE_NO_ADDRESS_SAFETY_ANALYSIS
 #endif
#endif
#ifdef WITH_PYMALLOC
#ifdef HAVE_MMAP
 #include <sys/mman.h>
 #ifdef MAP_ANONYMOUS
  #define ARENAS_USE_MMAP
 #endif
#endif
#ifdef WITH_VALGRIND
#include <valgrind/valgrind.h>
/* If we're using GCC, use __builtin_expect() to reduce overhead of
   the valgrind checks */
#if defined(__GNUC__) && (__GNUC__ > 2) && defined(__OPTIMIZE__)
#  define UNLIKELY(value) __builtin_expect((value), 0)
#else
#  define UNLIKELY(value) (value)
#endif
/* -1 indicates that we haven't checked that we're running on valgrind yet. */
static int running_on_valgrind = -1;
#endif
/* An object allocator for Python.
   Here is an introduction to the layers of the Python memory architecture,
   showing where the object allocator is actually used (layer +2), It is
   called for every object allocation and deallocation (PyObject_New/Del),
   unless the object-specific allocators implement a proprietary allocation
   scheme (ex.: ints use a simple free list). This is also the place where
   the cyclic garbage collector operates selectively on container objects.
 * -2层为物理存储器
 * -1层为操作系统的内存管理子系统
 *  0层为C语言库的内存分配器：例如malloc\free
 *  1层Python对0层的简单封装：主要解决标准C语言未定义清楚的情况，例如malloc(0)
 *  2层Python统一对象分配器：下面讲到的内存管理主要在这一层次
 *  3层Python的各对象分配器：每个类型自己管理类型对象的分配和回收，实际上就是缓存
    Object-specific allocators
    _____   ______   ______       ________
   [ int ] [ dict ] [ list ] ... [ string ]       Python core         |
+3 | <----- Object-specific memory -----> | <-- Non-object memory --> |
    _______________________________       |                           |
   [   Python's object allocator   ]      |                           |
+2 | ####### Object memory ####### | <------ Internal buffers ------> |
    ______________________________________________________________    |
   [          Python's raw memory allocator (PyMem_ API)          ]   |
+1 | <----- Python memory (under PyMem manager's control) ------> |   |
    __________________________________________________________________
   [    Underlying general-purpose allocator (ex: C library malloc)   ]
 0 | <------ Virtual memory allocated for the python process -------> |
   =========================================================================
    _______________________________________________________________________
   [                OS-specific Virtual Memory Manager (VMM)               ]
-1 | <--- Kernel dynamic storage allocation & management (page-based) ---> |
    __________________________________   __________________________________
   [                                  ] [                                  ]
-2 | <-- Physical memory: ROM/RAM --> | | <-- Secondary storage (swap) --> |
*/
/*==========================================================================*/
/* A fast, special-purpose memory allocator for small blocks, to be used
   on top of a general-purpose malloc -- heavily based on previous art. */
/* Vladimir Marangozov -- August 2000 */
/*
 * "Memory management is where the rubber meets the road -- if we do the wrong
 * thing at any level, the results will not be good. And if we don't make the
 * levels work well together, we are in serious trouble." (1)
 *
 * (1) Paul R. Wilson, Mark S. Johnstone, Michael Neely, and David Boles,
 *    "Dynamic Storage Allocation: A Survey and Critical Review",
 *    in Proc. 1995 Int'l. Workshop on Memory Management, September 1995.
 */
/* #undef WITH_MEMORY_LIMITS */         /* disable mem limit checks  */
/*==========================================================================*/
/*
 * Allocation strategy abstract:
 *
 * For small requests, the allocator sub-allocates <Big> blocks of memory.
 * Requests greater than SMALL_REQUEST_THRESHOLD bytes are routed to the
 * system's allocator. 
 *
 * Small requests are grouped in size classes spaced 8 bytes apart, due
 * to the required valid alignment of the returned address. Requests of
 * a particular size are serviced from memory pools of 4K (one VMM page).
 * Pools are fragmented on demand and contain free lists of blocks of one
 * particular size class. In other words, there is a fixed-size allocator
 * for each size class. Free pools are shared by the different allocators
 * thus minimizing the space reserved for a particular size class.
 *
 * This allocation strategy is a variant of what is known as "simple
 * segregated storage based on array of free lists". The main drawback of
 * simple segregated storage is that we might end up with lot of reserved
 * memory for the different free lists, which degenerate in time. To avoid
 * this, we partition each free list in pools and we share dynamically the
 * reserved space between all free lists. This technique is quite efficient
 * for memory intensive programs which allocate mainly small-sized blocks.
 *
 * 将内存按照大小分成(最多)64组，每8byte分为一组，每1个byte也叫1个block(uchar)。一个分配单元包含不等的block
 *
 * block: 1个block是一个字节(uchar)，内存分配的最小单位，只能分配整数个block的内存。实际上block只是一个抽象的概念或者说单位，没有实际的代码结构对应。
 *
 * 注意：这里面的size指请求的内存的大小nbytes；PyObject_Malloc代码中的size变量指请求的内存折算成的block个数(nblocks)。
 *
 * pool：1个pool包含一个pool_header和多个block。实际上就是pool的管理单元pool_header以及余下的可以分配出去的内存。同一个pool每次分配出去的内存都是固定数量的block，用szidx表示。szidx=1那么pool每次分配出16个block。1个pool实际上就是分配的一页4K内存，模型如下：
 *
 *  ---------------------------------------------------------
 * | pool_header  |用于对齐的废弃内存 |     可以分配的blocks|
 *  ---------------------------------------------------------
 *
 * For small requests we have the following table:
 *
 * Request in bytes     Size of allocated block      Size class idx
 * ----------------------------------------------------------------
 *        1-8                     8                       0
 *        9-16                   16                       1
 *       17-24                   24                       2
 *       25-32                   32                       3
 *       33-40                   40                       4
 *       41-48                   48                       5
 *       49-56                   56                       6
 *       57-64                   64                       7
 *       65-72                   72                       8
 *        ...                   ...                     ...
 *      497-504                 504                      62
 *      505-512                 512                      63 
 *
 *      0, SMALL_REQUEST_THRESHOLD + 1 and up: routed to the underlying
 *      allocator.
 */
/*==========================================================================*/
/*
 * -- Main tunable settings section --
 */
/*
 * Alignment of addresses returned to the user. 8-bytes alignment works
 * on most current architectures (with 32-bit or 64-bit address busses).
 * The alignment value is also used for grouping small requests in size
 * classes spaced ALIGNMENT bytes apart.
 *
 * You shouldn't change this unless you know what you are doing.
 */
#define ALIGNMENT               8               /* must be 2^N */
#define ALIGNMENT_SHIFT         3
#define ALIGNMENT_MASK          (ALIGNMENT - 1)
/* Return the number of bytes in size class I, as a uint. */
#define INDEX2SIZE(I) (((uint)(I) + 1) << ALIGNMENT_SHIFT)
/*
 * Max size threshold below which malloc requests are considered to be
 * small enough in order to use preallocated memory pools. You can tune
 * this value according to your application behaviour and memory needs.
 *
 * The following invariants must hold:
 *      1) ALIGNMENT <= SMALL_REQUEST_THRESHOLD <= 512
 *      2) SMALL_REQUEST_THRESHOLD is evenly divisible by ALIGNMENT
 *
 * Note: a size threshold of 512 guarantees that newly created dictionaries
 * will be allocated from preallocated memory pools on 64-bit.
 *
 * Although not required, for better performance and space efficiency,
 * it is recommended that SMALL_REQUEST_THRESHOLD is set to a power of 2.
 *
 * SMALL_REQUEST_THRESHOLD必须大于等于ALIGNMENT，不然内存管理无意义；
 * 必须小于等于512 是因为usedpools代码写死最多支持64个不同单位szidx的pools。
 */
#define SMALL_REQUEST_THRESHOLD 512 
#define NB_SMALL_SIZE_CLASSES   (SMALL_REQUEST_THRESHOLD / ALIGNMENT)
/*
 * The system's VMM page size can be obtained on most unices with a
 * getpagesize() call or deduced from various header files. To make
 * things simpler, we assume that it is 4K, which is OK for most systems.
 * It is probably better if this is the native page size, but it doesn't
 * have to be.  In theory, if SYSTEM_PAGE_SIZE is larger than the native page
 * size, then `POOL_ADDR(p)->arenaindex' could rarely cause a segmentation
 * violation fault.  4K is apparently OK for all the platforms that python
 * currently targets.
 */
#define SYSTEM_PAGE_SIZE        (4 * 1024)
#define SYSTEM_PAGE_SIZE_MASK   (SYSTEM_PAGE_SIZE - 1)
/*
 * Maximum amount of memory managed by the allocator for small requests.
 */
#ifdef WITH_MEMORY_LIMITS
#ifndef SMALL_MEMORY_LIMIT
#define SMALL_MEMORY_LIMIT      (64 * 1024 * 1024)      /* 64 MB -- more? */
#endif
#endif
/*
 * The allocator sub-allocates <Big> blocks of memory (called arenas) aligned
 * on a page boundary. This is a reserved virtual address space for the
 * current process (obtained through a malloc()/mmap() call). In no way this
 * means that the memory arenas will be used entirely. A malloc(<Big>) is
 * usually an address range reservation for <Big> bytes, unless all pages within
 * this space are referenced subsequently. So malloc'ing big blocks and not
 * using them does not mean "wasting memory". It's an addressable range
 * wastage... 
 *
 * Arenas are allocated with mmap() on systems supporting anonymous memory
 * mappings to reduce heap fragmentation.
 *
 * arena：1个arena为256KB，每个arena中包含多个pool。1个arena中的pool(1个页)的szidx可以不同。
 * arena有3种组织结构：
 * 1. arenas数组，arena_object数组，arena_object.address是从系统请求的ARENA_SIZE内存的地址
 * 2. unused_arena_objects单向链表，链接所有空闲的arena
 * 3. usable_arenas双向链表，链接所有部分内存被分配出去的arena
 */
#define ARENA_SIZE              (256 << 10)     /* 256KB */
#ifdef WITH_MEMORY_LIMITS
#define MAX_ARENAS              (SMALL_MEMORY_LIMIT / ARENA_SIZE)
#endif
/*
 * Size of the pools used for small blocks. Should be a power of 2,
 * between 1K and SYSTEM_PAGE_SIZE, that is: 1k, 2k, 4k.
 *
 * POOL的大小，一般为4K
 */
#define POOL_SIZE               SYSTEM_PAGE_SIZE        /* must be 2^N */
#define POOL_SIZE_MASK          SYSTEM_PAGE_SIZE_MASK
/*
 * -- End of tunable settings section --
 */
/*==========================================================================*/
/*
 * Locking
 *
 * To reduce lock contention, it would probably be better to refine the
 * crude function locking with per size class locking. I'm not positive
 * however, whether it's worth switching to such locking policy because
 * of the performance penalty it might introduce.
 *
 * The following macros describe the simplest (should also be the fastest)
 * lock object on a particular platform and the init/fini/lock/unlock
 * operations on it. The locks defined here are not expected to be recursive
 * because it is assumed that they will always be called in the order:
 * INIT, [LOCK, UNLOCK]*, FINI.
 */
/*
 * Python's threads are serialized, so object malloc locking is disabled.
 */
#define SIMPLELOCK_DECL(lock)   /* simple lock declaration              */
#define SIMPLELOCK_INIT(lock)   /* allocate (if needed) and initialize  */
#define SIMPLELOCK_FINI(lock)   /* free/destroy an existing lock        */
#define SIMPLELOCK_LOCK(lock)   /* acquire released lock */
#define SIMPLELOCK_UNLOCK(lock) /* release acquired lock */
/*
 * Basic types
 * I don't care if these are defined in <sys/types.h> or elsewhere. Axiom.
 */
#undef  uchar
#define uchar   unsigned char   /* assuming == 8 bits  */
#undef  uint
#define uint    unsigned int    /* assuming >= 16 bits */
#undef  ulong
#define ulong   unsigned long   /* assuming >= 32 bits */
#undef uptr
#define uptr    Py_uintptr_t
/* When you say memory, my mind reasons in terms of (pointers to) blocks */
typedef uchar block;
/* Pool for small blocks. */
struct pool_header {
    /* 用于对齐，使一个Pool的2个block处是nextpool。
     * 当pool回收时可以加快再分配 */
    union { block *_padding;
            uint count; } ref;          /* number of allocated blocks    */
    /* 1个Pool内未被分配出去的/回收的内存的链表 */
    block *freeblock;                   /* pool's free list head         */
    /* 链接相同szidx的pool，在不同的链表下作用不同 */
    struct pool_header *nextpool;       /* next pool of this size class  */
    struct pool_header *prevpool;       /* previous pool       ""        */
    /* 所处的arena在arenas数组中的小标
     * 当释放内存时用来判断给定的内存地址是否是由Pool分配出去的*/
    uint arenaindex;                    /* index into arenas of base adr */
    /* Pool每次只能分配指定szidx的blocks个内存 */
    uint szidx;                         /* block size class index        */
    /* 从来没有被分配出去的连续的剩余内存的偏移地址 */
    uint nextoffset;                    /* bytes to virgin block         */
    /* 可用内存的最大偏移量
    uint maxnextoffset;                 /* largest valid nextoffset      */
};
typedef struct pool_header *poolp;
/* Record keeping for arenas. */
struct arena_object {
    /* The address of the arena, as returned by malloc.  Note that 0
     * will never be returned by a successful malloc, and is used
     * here to mark an arena_object that doesn't correspond to an
     * allocated arena.
     */
    /*
     * 一个arena的内存起始值256KB
     */
    uptr address;
    /* Pool-aligned pointer to the next pool to be carved off. */
    /* 剩余的连续地址的pool的地址 */
    block* pool_address;
    /* The number of available pools in the arena:  free pools + never-
     * allocated pools.
     */
    /* 空闲的Pool数量, 被回收的pool的链表 */
    uint nfreepools;
    /* The total number of pools in the arena, whether or not available. */
    uint ntotalpools;
    /* Singly-linked list of available pools. */
    /* 空闲Pool的链表 */
    struct pool_header* freepools;
    /* Whenever this arena_object is not associated with an allocated
     * arena, the nextarena member is used to link all unassociated
     * arena_objects in the singly-linked `unused_arena_objects` list.
     * The prevarena member is unused in this case.
     *
     * When this arena_object is associated with an allocated arena
     * with at least one available pool, both members are used in the
     * doubly-linked `usable_arenas` list, which is maintained in
     * increasing order of `nfreepools` values.
     *
     * Else this arena_object is associated with an allocated arena
     * all of whose pools are in use.  `nextarena` and `prevarena`
     * are both meaningless in this case.
     */
    /* arena还有两种组织结构：
     * 第一种：unused_arena_objects
     * 第二种：usable_arenas 
    struct arena_object* nextarena;
    struct arena_object* prevarena;
};
#undef  ROUNDUP
#define ROUNDUP(x)              (((x) + ALIGNMENT_MASK) & ~ALIGNMENT_MASK)
#define POOL_OVERHEAD           ROUNDUP(sizeof(struct pool_header))
#define DUMMY_SIZE_IDX          0xffff  /* size class of newly cached pools */
/* Round pointer P down to the closest pool-aligned address <= P, as a poolp */
/* 给定free的内存地址，向上对齐找到所属的POOL */
#define POOL_ADDR(P) ((poolp)((uptr)(P) & ~(uptr)POOL_SIZE_MASK))
/* Return total number of blocks in pool of size index I, as a uint. */
/* POOL中有多少个block，(POOL_SIZE - pool_object的大小) / 分配单位的大小)
#define NUMBLOCKS(I) ((uint)(POOL_SIZE - POOL_OVERHEAD) / INDEX2SIZE(I))
/*==========================================================================*/
/*
 * This malloc lock
 */
SIMPLELOCK_DECL(_malloc_lock)
#define LOCK()          SIMPLELOCK_LOCK(_malloc_lock)
#define UNLOCK()        SIMPLELOCK_UNLOCK(_malloc_lock)
#define LOCK_INIT()     SIMPLELOCK_INIT(_malloc_lock)
#define LOCK_FINI()     SIMPLELOCK_FINI(_malloc_lock)
/*
 * Pool table -- headed, circular, doubly-linked lists of partially used pools.
This is involved.  For an index i, usedpools[i+i] is the header for a list of
all partially used pools holding small blocks with "size class idx" i. So
usedpools[0] corresponds to blocks of size 8, usedpools[2] to blocks of size
16, and so on:  index 2*i <-> blocks of size (i+1)<<ALIGNMENT_SHIFT.
Pools are carved off an arena's highwater mark (an arena_object's pool_address
member) as needed.  Once carved off, a pool is in one of three states forever
after:
used == partially used, neither empty nor full
    At least one block in the pool is currently allocated, and at least one
    block in the pool is not currently allocated (note this implies a pool
    has room for at least two blocks).
    This is a pool's initial state, as a pool is created only when malloc
    needs space.
    The pool holds blocks of a fixed size, and is in the circular list headed
    at usedpools[i] (see above).  It's linked to the other used pools of the
    same size class via the pool_header's nextpool and prevpool members.
    If all but one block is currently allocated, a malloc can cause a
    transition to the full state.  If all but one block is not currently
    allocated, a free can cause a transition to the empty state.
full == all the pool's blocks are currently allocated
    On transition to full, a pool is unlinked from its usedpools[] list.
    It's not linked to from anything then anymore, and its nextpool and
    prevpool members are meaningless until it transitions back to used.
    A free of a block in a full pool puts the pool back in the used state.
    Then it's linked in at the front of the appropriate usedpools[] list, so
    that the next allocation for its size class will reuse the freed block.
empty == all the pool's blocks are currently available for allocation
    On transition to empty, a pool is unlinked from its usedpools[] list,
    and linked to the front of its arena_object's singly-linked freepools list,
    via its nextpool member.  The prevpool member has no meaning in this case.
    Empty pools have no inherent size class:  the next time a malloc finds
    an empty list in usedpools[], it takes the first pool off of freepools.
    If the size class needed happens to be the same as the size class the pool
    last had, some pool initialization can be skipped.
Block Management
Blocks within pools are again carved out as needed.  pool->freeblock points to
the start of a singly-linked list of free blocks within the pool.  When a
block is freed, it's inserted at the front of its pool's freeblock list.  Note
that the available blocks in a pool are *not* linked all together when a pool
is initialized.  Instead only "the first two" (lowest addresses) blocks are
set up, returning the first such block, and setting pool->freeblock to a
one-block list holding the second such block.  This is consistent with that
pymalloc strives at all levels (arena, pool, and block) never to touch a piece
of memory until it's actually needed.
So long as a pool is in the used state, we're certain there *is* a block
available for allocating, and pool->freeblock is not NULL.  If pool->freeblock
points to the end of the free list before we've carved the entire pool into
blocks, that means we simply haven't yet gotten to one of the higher-address
blocks.  The offset from the pool_header to the start of "the next" virgin
block is stored in the pool_header nextoffset member, and the largest value
of nextoffset that makes sense is stored in the maxnextoffset member when a
pool is initialized.  All the blocks in a pool have been passed out at least
once when and only when nextoffset > maxnextoffset.
Major obscurity:  While the usedpools vector is declared to have poolp
entries, it doesn't really.  It really contains two pointers per (conceptual)
poolp entry, the nextpool and prevpool members of a pool_header.  The
excruciating initialization code below fools C so that
    usedpool[i+i]
"acts like" a genuine poolp, but only so long as you only reference its
nextpool and prevpool members.  The "- 2*sizeof(block *)" gibberish is
compensating for that a pool_header's nextpool and prevpool members
immediately follow a pool_header's first two members:
    union { block *_padding;
            uint count; } ref;
    block *freeblock;
each of which consume sizeof(block *) bytes.  So what usedpools[i+i] really
contains is a fudged-up pointer p such that *if* C believes it's a poolp
pointer, then p->nextpool and p->prevpool are both p (meaning that the headed
circular list is empty).
It's unclear why the usedpools setup is so convoluted.  It could be to
minimize the amount of cache required to hold this heavily-referenced table
(which only *needs* the two interpool pointer members of a pool_header). OTOH,
referencing code has to remember to "double the index" and doing so isn't
free, usedpools[0] isn't a strictly legal pointer, and we're crucially relying
on that C doesn't insert any padding anywhere in a pool_header at or before
the prevpool member.
**************************************************************************** */
#define PTA(x)  ((poolp )((uchar *)&(usedpools[2*(x)]) - 2*sizeof(block *)))
#define PT(x)   PTA(x), PTA(x)
static poolp usedpools[2 * ((NB_SMALL_SIZE_CLASSES + 7) / 8) * 8] = {
    PT(0), PT(1), PT(2), PT(3), PT(4), PT(5), PT(6), PT(7)
#if NB_SMALL_SIZE_CLASSES > 8
    , PT(8), PT(9), PT(10), PT(11), PT(12), PT(13), PT(14), PT(15)
#if NB_SMALL_SIZE_CLASSES > 16
    , PT(16), PT(17), PT(18), PT(19), PT(20), PT(21), PT(22), PT(23)
#if NB_SMALL_SIZE_CLASSES > 24
    , PT(24), PT(25), PT(26), PT(27), PT(28), PT(29), PT(30), PT(31)
#if NB_SMALL_SIZE_CLASSES > 32
    , PT(32), PT(33), PT(34), PT(35), PT(36), PT(37), PT(38), PT(39)
#if NB_SMALL_SIZE_CLASSES > 40
    , PT(40), PT(41), PT(42), PT(43), PT(44), PT(45), PT(46), PT(47)
#if NB_SMALL_SIZE_CLASSES > 48
    , PT(48), PT(49), PT(50), PT(51), PT(52), PT(53), PT(54), PT(55)
#if NB_SMALL_SIZE_CLASSES > 56
    , PT(56), PT(57), PT(58), PT(59), PT(60), PT(61), PT(62), PT(63)
#if NB_SMALL_SIZE_CLASSES > 64
#error "NB_SMALL_SIZE_CLASSES should be less than 64"
#endif /* NB_SMALL_SIZE_CLASSES > 64 */
#endif /* NB_SMALL_SIZE_CLASSES > 56 */
#endif /* NB_SMALL_SIZE_CLASSES > 48 */
#endif /* NB_SMALL_SIZE_CLASSES > 40 */
#endif /* NB_SMALL_SIZE_CLASSES > 32 */
#endif /* NB_SMALL_SIZE_CLASSES > 24 */
#endif /* NB_SMALL_SIZE_CLASSES > 16 */
#endif /* NB_SMALL_SIZE_CLASSES >  8 */
};
/*==========================================================================
Arena management.
`arenas` is a vector of arena_objects.  It contains maxarenas entries, some of
which may not be currently used (== they're arena_objects that aren't
currently associated with an allocated arena).  Note that arenas proper are
separately malloc'ed.
Prior to Python 2.5, arenas were never free()'ed.  Starting with Python 2.5,
we do try to free() arenas, and use some mild heuristic strategies to increase
the likelihood that arenas eventually can be freed.
unused_arena_objects
    This is a singly-linked list of the arena_objects that are currently not
    being used (no arena is associated with them).  Objects are taken off the
    head of the list in new_arena(), and are pushed on the head of the list in
    PyObject_Free() when the arena is empty.  Key invariant:  an arena_object
    is on this list if and only if its .address member is 0.
    arena_object的单向链表，用.nextarena链接，.address为0
    刚从new_arena分配或者空闲的时候会加入该链表
usable_arenas
    This is a doubly-linked list of the arena_objects associated with arenas
    that have pools available.  These pools are either waiting to be reused,
    or have not been used before.  The list is sorted to have the most-
    allocated arenas first (ascending order based on the nfreepools member).
    This means that the next allocation will come from a heavily used arena,
    which gives the nearly empty arenas a chance to be returned to the system.
    In my unscientific tests this dramatically improved the number of arenas
    that could be freed.
    arena_object的双向链表，并且arena_object中有可用的pools。
    已分配出内存最多的arena排在前面，以便最空闲的arena有机会因为内存都被回收而被回收。
Note that an arena_object associated with an arena all of whose pools are
currently in use isn't on either list.
注意所有pools都被使用了的arena不会在这两个链表中。
*/
/* Array of objects used to track chunks of memory (arenas). */
static struct arena_object* arenas = NULL;
/* Number of slots currently allocated in the `arenas` vector. */
static uint maxarenas = 0;
/* The head of the singly-linked, NULL-terminated list of available
 * arena_objects.
 */
static struct arena_object* unused_arena_objects = NULL;
/* The head of the doubly-linked, NULL-terminated at each end, list of
 * arena_objects associated with arenas that have pools available.
 */
static struct arena_object* usable_arenas = NULL;
/* How many arena_objects do we initially allocate?
 * 16 = can allocate 16 arenas = 16 * ARENA_SIZE = 4MB before growing the
 * `arenas` vector.
 */
#define INITIAL_ARENA_OBJECTS 16
/* Number of arenas allocated that haven't been free()'d. */
static size_t narenas_currently_allocated = 0;
#ifdef PYMALLOC_DEBUG
/* Total number of times malloc() called to allocate an arena. */
static size_t ntimes_arena_allocated = 0;
/* High water mark (max value ever seen) for narenas_currently_allocated. */
static size_t narenas_highwater = 0;
#endif
/* Allocate a new arena.  If we run out of memory, return NULL.  Else
 * allocate a new arena, and return the address of an arena_object
 * describing the new arena.  It's expected that the caller will set
 * `usable_arenas` to the return value.
 */
static struct arena_object*
new_arena(void)
{
    struct arena_object* arenaobj;
    uint excess;        /* number of bytes above pool alignment */
    void *address;
    int err;
#ifdef PYMALLOC_DEBUG
    if (Py_GETENV("PYTHONMALLOCSTATS"))
        _PyObject_DebugMallocStats();
#endif
    /* Python初始化，或者所有的arenas都耗尽了 */
    if (unused_arena_objects == NULL) {
        uint i;
        uint numarenas;
        size_t nbytes;
        /* Double the number of arena objects on each allocation.
         * Note that it's possible for `numarenas` to overflow.
         */
        /* 每次分配arena的数量倍增 16 -> 32 */
        numarenas = maxarenas ? maxarenas << 1 : INITIAL_ARENA_OBJECTS;
        /* 溢出导致不能再分配arena了 */
        if (numarenas <= maxarenas)
            return NULL;                /* overflow */
#if SIZEOF_SIZE_T <= SIZEOF_INT
        if (numarenas > PY_SIZE_MAX / sizeof(*arenas))
            return NULL;                /* overflow */
#endif
        nbytes = numarenas * sizeof(*arenas);
        arenaobj = (struct arena_object *)realloc(arenas, nbytes);
        if (arenaobj == NULL)
            return NULL;
        arenas = arenaobj;
        /* We might need to fix pointers that were copied.  However,
         * new_arena only gets called when all the pages in the
         * previous arenas are full.  Thus, there are *no* pointers
         * into the old array. Thus, we don't have to worry about
         * invalid pointers.  Just to be sure, some asserts:
         */
        /* usable_arenas 和 unused_arena_objects 都为空。
         * 所有的pools都分配出去了，才会导致申请新的arena
         */ 
        assert(usable_arenas == NULL);
        assert(unused_arena_objects == NULL);
        /* Put the new arenas on the unused_arena_objects list. */
        for (i = maxarenas; i < numarenas; ++i) {
            arenas[i].address = 0;              /* mark as unassociated */
            arenas[i].nextarena = i < numarenas - 1 ?
                                   &arenas[i+1] : NULL;
        }
        /* Update globals. */
        unused_arena_objects = &arenas[maxarenas];
        maxarenas = numarenas;
    }
    /* Take the next available arena object off the head of the list. */
    assert(unused_arena_objects != NULL);
    arenaobj = unused_arena_objects;
    unused_arena_objects = arenaobj->nextarena;
    assert(arenaobj->address == 0);
    /* 分配1个arena的内存 64个POOL */
#ifdef ARENAS_USE_MMAP
    address = mmap(NULL, ARENA_SIZE, PROT_READ|PROT_WRITE,
                   MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
    err = (address == MAP_FAILED);
#else
    address = malloc(ARENA_SIZE);
    err = (address == 0);
#endif    
    if (err) {
        /* The allocation failed: return NULL after putting the
         * arenaobj back.
         */
        arenaobj->nextarena = unused_arena_objects;
        unused_arena_objects = arenaobj;
        return NULL;
    }
    /* arena中的address就是分配的256KB内存的地址 */
    arenaobj->address = (uptr)address;
    ++narenas_currently_allocated;
#ifdef PYMALLOC_DEBUG
    ++ntimes_arena_allocated;
    if (narenas_currently_allocated > narenas_highwater)
        narenas_highwater = narenas_currently_allocated;
#endif
    /* 注意这里的freepools =  NULL */
    arenaobj->freepools = NULL;
    /* pool_address <- first pool-aligned address in the arena
       nfreepools <- number of whole pools that fit after alignment */
    arenaobj->pool_address = (block*)arenaobj->address;
    arenaobj->nfreepools = ARENA_SIZE / POOL_SIZE;
    assert(POOL_SIZE * arenaobj->nfreepools == ARENA_SIZE);
    /* 将arena的内存按照Pool对齐 */
    excess = (uint)(arenaobj->address & POOL_SIZE_MASK);
    if (excess != 0) {
        --arenaobj->nfreepools;
        arenaobj->pool_address += POOL_SIZE - excess;
    }
    arenaobj->ntotalpools = arenaobj->nfreepools;
    return arenaobj;
}
/*
Py_ADDRESS_IN_RANGE(P, POOL)
Return true if and only if P is an address that was allocated by pymalloc.
POOL must be the pool address associated with P, i.e., POOL = POOL_ADDR(P)
(the caller is asked to compute this because the macro expands POOL more than
once, and for efficiency it's best for the caller to assign POOL_ADDR(P) to a
variable and pass the latter to the macro; because Py_ADDRESS_IN_RANGE is
called on every alloc/realloc/free, micro-efficiency is important here).
Tricky:  Let B be the arena base address associated with the pool, B =
arenas[(POOL)->arenaindex].address.  Then P belongs to the arena if and only if
    B <= P < B + ARENA_SIZE
Subtracting B throughout, this is true iff
    0 <= P-B < ARENA_SIZE
By using unsigned arithmetic, the "0 <=" half of the test can be skipped.
Obscure:  A PyMem "free memory" function can call the pymalloc free or realloc
before the first arena has been allocated.  `arenas` is still NULL in that
case.  We're relying on that maxarenas is also 0 in that case, so that
(POOL)->arenaindex < maxarenas  must be false, saving us from trying to index
into a NULL arenas.
Details:  given P and POOL, the arena_object corresponding to P is AO =
arenas[(POOL)->arenaindex].  Suppose obmalloc controls P.  Then (barring wild
stores, etc), POOL is the correct address of P's pool, AO.address is the
correct base address of the pool's arena, and P must be within ARENA_SIZE of
AO.address.  In addition, AO.address is not 0 (no arena can start at address 0
(NULL)).  Therefore Py_ADDRESS_IN_RANGE correctly reports that obmalloc
controls P.
Now suppose obmalloc does not control P (e.g., P was obtained via a direct
call to the system malloc() or realloc()).  (POOL)->arenaindex may be anything
in this case -- it may even be uninitialized trash.  If the trash arenaindex
is >= maxarenas, the macro correctly concludes at once that obmalloc doesn't
control P.
Else arenaindex is < maxarena, and AO is read up.  If AO corresponds to an
allocated arena, obmalloc controls all the memory in slice AO.address :
AO.address+ARENA_SIZE.  By case assumption, P is not controlled by obmalloc,
so P doesn't lie in that slice, so the macro correctly reports that P is not
controlled by obmalloc.
Finally, if P is not controlled by obmalloc and AO corresponds to an unused
arena_object (one not currently associated with an allocated arena),
AO.address is 0, and the second test in the macro reduces to:
    P < ARENA_SIZE
If P >= ARENA_SIZE (extremely likely), the macro again correctly concludes
that P is not controlled by obmalloc.  However, if P < ARENA_SIZE, this part
of the test still passes, and the third clause (AO.address != 0) is necessary
to get the correct result:  AO.address is 0 in this case, so the macro
correctly reports that P is not controlled by obmalloc (despite that P lies in
slice AO.address : AO.address + ARENA_SIZE).
Note:  The third (AO.address != 0) clause was added in Python 2.5.  Before
2.5, arenas were never free()'ed, and an arenaindex < maxarena always
corresponded to a currently-allocated arena, so the "P is not controlled by
obmalloc, AO corresponds to an unused arena_object, and P < ARENA_SIZE" case
was impossible.
Note that the logic is excruciating, and reading up possibly uninitialized
memory when P is not controlled by obmalloc (to get at (POOL)->arenaindex)
creates problems for some memory debuggers.  The overwhelming advantage is
that this test determines whether an arbitrary address is controlled by
obmalloc in a small constant time, independent of the number of arenas
obmalloc controls.  Since this test is needed at every entry point, it's
extremely desirable that it be this fast.
Since Py_ADDRESS_IN_RANGE may be reading from memory which was not allocated
by Python, it is important that (POOL)->arenaindex is read only once, as
another thread may be concurrently modifying the value without holding the
GIL.  To accomplish this, the arenaindex_temp variable is used to store
(POOL)->arenaindex for the duration of the Py_ADDRESS_IN_RANGE macro's
execution.  The caller of the macro is responsible for declaring this
variable.
*/
#define Py_ADDRESS_IN_RANGE(P, POOL)                    \
    ((arenaindex_temp = (POOL)->arenaindex) < maxarenas &&              \
     (uptr)(P) - arenas[arenaindex_temp].address < (uptr)ARENA_SIZE && \
     arenas[arenaindex_temp].address != 0)
/* This is only useful when running memory debuggers such as
 * Purify or Valgrind.  Uncomment to use.
 *
#define Py_USING_MEMORY_DEBUGGER
 */
#ifdef Py_USING_MEMORY_DEBUGGER
/* Py_ADDRESS_IN_RANGE may access uninitialized memory by design
 * This leads to thousands of spurious warnings when using
 * Purify or Valgrind.  By making a function, we can easily
 * suppress the uninitialized memory reads in this one function.
 * So we won't ignore real errors elsewhere.
 *
 * Disable the macro and use a function.
 */
#undef Py_ADDRESS_IN_RANGE
#if defined(__GNUC__) && ((__GNUC__ == 3) && (__GNUC_MINOR__ >= 1) || \
                          (__GNUC__ >= 4))
#define Py_NO_INLINE __attribute__((__noinline__))
#else
#define Py_NO_INLINE
#endif
/* Don't make static, to try to ensure this isn't inlined. */
int Py_ADDRESS_IN_RANGE(void *P, poolp pool) Py_NO_INLINE;
#undef Py_NO_INLINE
#endif
/*==========================================================================*/
/* malloc.  Note that nbytes==0 tries to return a non-NULL pointer, distinct
 * from all other currently live pointers.  This may not be possible.
 */
/*
 * The basic blocks are ordered by decreasing execution frequency,
 * which minimizes the number of jumps in the most common cases,
 * improves branching prediction and instruction scheduling (small
 * block allocations typically result in a couple of instructions).
 * Unless the optimizer reorders everything, being too smart...
 */
#undef PyObject_Malloc
void *
PyObject_Malloc(size_t nbytes)
{
    block *bp;
    poolp pool;
    poolp next;
    uint size;
#ifdef WITH_VALGRIND
    if (UNLIKELY(running_on_valgrind == -1))
        running_on_valgrind = RUNNING_ON_VALGRIND;
    if (UNLIKELY(running_on_valgrind))
        goto redirect;
#endif
    /*
     * Limit ourselves to PY_SSIZE_T_MAX bytes to prevent security holes.
     * Most python internals blindly use a signed Py_ssize_t to track
     * things without checking for overflows or negatives.
     * As size_t is unsigned, checking for nbytes < 0 is not required.
     */
    if (nbytes > PY_SSIZE_T_MAX)
        return NULL;
    /*
     * This implicitly redirects malloc(0).
     */
    /* 等同于 0 < nbytes <= SMALL_REQUEST_THRESHOLD */
    if ((nbytes - 1) < SMALL_REQUEST_THRESHOLD) {
        LOCK();
        /*
         * Most frequent paths first
         */
        size = (uint)(nbytes - 1) >> ALIGNMENT_SHIFT;
        pool = usedpools[size + size];
        /* usedpools是一个环状链表 */
        if (pool != pool->nextpool) {
            /*
             * There is a used pool for this size class.
             * Pick up the head block of its free list.
             */
            ++pool->ref.count;
            /* freeblock 是一个链表。freeblock是头，
             * 每个指针存储在可分配单元的第一个block中 
             */
            bp = pool->freeblock;
            assert(bp != NULL);
            /* freeblock不为空 */
            if ((pool->freeblock = *(block **)bp) != NULL) {
                UNLOCK();
                return (void *)bp;
            }
            /*
             * Reached the end of the free list, try to extend it.
             */
            /* 至少还有1个size的block可供分配 */
            if (pool->nextoffset <= pool->maxnextoffset) {
                /* There is room for another block. */
                pool->freeblock = (block*)pool +
                                  pool->nextoffset;
                pool->nextoffset += INDEX2SIZE(size);
                *(block **)(pool->freeblock) = NULL;
                UNLOCK();
                return (void *)bp;
            }
            /* Pool is full, unlink from used pools. */
            /* Pool都分配出去了，从used pools中拆除去 */
            next = pool->nextpool;
            pool = pool->prevpool;
            next->prevpool = pool;
            pool->nextpool = next;
            UNLOCK();
            return (void *)bp;
        }
        /* There isn't a pool of the right size class immediately
         * available:  use a free pool.
         */
        /* 没有空闲的pool，也没有空闲的arena */
        if (usable_arenas == NULL) {
            /* No arena has a free pool:  allocate a new arena. */
#ifdef WITH_MEMORY_LIMITS
            if (narenas_currently_allocated >= MAX_ARENAS) {
                UNLOCK();
                goto redirect;
            }
#endif
            usable_arenas = new_arena();
            if (usable_arenas == NULL) {
                UNLOCK();
                goto redirect;
            }
            usable_arenas->nextarena =
                usable_arenas->prevarena = NULL;
        }
        assert(usable_arenas->address != 0);
        /* Try to get a cached free pool. */
        pool = usable_arenas->freepools;
        /* arena不是新分配的，新分配的arena.freepools == NULL */
        if (pool != NULL) {
            /* Unlink from cached pools. */
            usable_arenas->freepools = pool->nextpool;
            /* This arena already had the smallest nfreepools
             * value, so decreasing nfreepools doesn't change
             * that, and we don't need to rearrange the
             * usable_arenas list.  However, if the arena has
             * become wholly allocated, we need to remove its
             * arena_object from usable_arenas.
             */
            /* arena都分配出去了，从usable_arenas中拆除 */
            --usable_arenas->nfreepools;
            if (usable_arenas->nfreepools == 0) {
                /* Wholly allocated:  remove. */
                assert(usable_arenas->freepools == NULL);
                assert(usable_arenas->nextarena == NULL ||
                       usable_arenas->nextarena->prevarena ==
                       usable_arenas);
                usable_arenas = usable_arenas->nextarena;
                if (usable_arenas != NULL) {
                    usable_arenas->prevarena = NULL;
                    assert(usable_arenas->address != 0);
                }
            }
            else {
                /* nfreepools > 0:  it must be that freepools
                 * isn't NULL, or that we haven't yet carved
                 * off all the arena's pools for the first
                 * time.
                 */
                assert(usable_arenas->freepools != NULL ||
                       usable_arenas->pool_address <=
                       (block*)usable_arenas->address +
                           ARENA_SIZE - POOL_SIZE);
            }
        init_pool:
            /* Frontlink to used pools. */
            next = usedpools[size + size]; /* == prev */
            pool->nextpool = next;
            pool->prevpool = next;
            next->nextpool = pool;
            next->prevpool = pool;
            pool->ref.count = 1;
            /* 正好回收的pool的szidx和这次用于分配的size一样 
             * 不需要初始化了
             */
            if (pool->szidx == size) {
                /* Luckily, this pool last contained blocks
                 * of the same size class, so its header
                 * and free list are already initialized.
                 */
                bp = pool->freeblock;
                pool->freeblock = *(block **)bp;
                UNLOCK();
                return (void *)bp;
            }
            /*
             * Initialize the pool header, set up the free list to
             * contain just the second block, and return the first
             * block.
             */
            pool->szidx = size;
            size = INDEX2SIZE(size);
            bp = (block *)pool + POOL_OVERHEAD;
            pool->nextoffset = POOL_OVERHEAD + (size << 1);
            pool->maxnextoffset = POOL_SIZE - size;
            pool->freeblock = bp + size;
            *(block **)(pool->freeblock) = NULL;
            UNLOCK();
            return (void *)bp;
        }
        /* Carve off a new pool. */
        /* 从usable_arenas中找到一个空闲的Pool */
        assert(usable_arenas->nfreepools > 0);
        assert(usable_arenas->freepools == NULL);
        /* pool_address空闲的POOL的地址 */
        pool = (poolp)usable_arenas->pool_address;
        assert((block*)pool <= (block*)usable_arenas->address +
                               ARENA_SIZE - POOL_SIZE);
        /* pool 所在的arena在arenas的下标就是当前usable_arenas相对arenas的偏移 */
        pool->arenaindex = usable_arenas - arenas;
        assert(&arenas[pool->arenaindex] == usable_arenas);
        /* 0xFFFF */
        pool->szidx = DUMMY_SIZE_IDX;
        usable_arenas->pool_address += POOL_SIZE;
        --usable_arenas->nfreepools;
        if (usable_arenas->nfreepools == 0) {
            /* usable_arenas中的arena是双向链表 */
            assert(usable_arenas->nextarena == NULL ||
                   usable_arenas->nextarena->prevarena ==
                   usable_arenas);
            /* Unlink the arena:  it is completely allocated. */
            usable_arenas = usable_arenas->nextarena;
            if (usable_arenas != NULL) {
                usable_arenas->prevarena = NULL;
                assert(usable_arenas->address != 0);
            }
        }
        goto init_pool;
    }
    /* The small block allocator ends here. */
redirect:
    /* Redirect the original request to the underlying (libc) allocator.
     * We jump here on bigger requests, on error in the code above (as a
     * last chance to serve the request) or when the max memory limit
     * has been reached.
     */
    if (nbytes == 0)
        nbytes = 1;
    return (void *)malloc(nbytes);
}
/* free */
#undef PyObject_Free
ATTRIBUTE_NO_ADDRESS_SAFETY_ANALYSIS
void
PyObject_Free(void *p)
{
    poolp pool;
    block *lastfree;
    poolp next, prev;
    uint size;
#ifndef Py_USING_MEMORY_DEBUGGER
    uint arenaindex_temp;
#endif
    if (p == NULL)      /* free(NULL) has no effect */
        return;
#ifdef WITH_VALGRIND
    if (UNLIKELY(running_on_valgrind > 0))
        goto redirect;
#endif
    /* 向上对齐4K，找到所在的Pool的地址 */
    pool = POOL_ADDR(p);
    /* p是通过pool分配出去的，只需证明p在pool所在的arena的地址内:
     * 1. pool->arenaindex <= maxarenas 确实至少分配了这么多arena
     * 2. 0 < p <= arenas[pool->arenaindex].address + ARENA_SIZE
     */
    if (Py_ADDRESS_IN_RANGE(p, pool)) {
        /* We allocated this address. */
        LOCK();
        /* Link p to the start of the pool's freeblock list.  Since
         * the pool had at least the p block outstanding, the pool
         * wasn't empty (so it's already in a usedpools[] list, or
         * was full and is in no list -- it's not in the freeblocks
         * list in any case).
         */
        assert(pool->ref.count > 0);            /* else it was empty */
        /* 将p放入freeblock链表头部 
         * 将前一个链表的头元素地址放入p的内存中
         */
        *(block **)p = lastfree = pool->freeblock;
        pool->freeblock = (block *)p;
        /* pool现在至少有2个分配单元: lastfree不为空1个，刚放入的1个 */
        if (lastfree) {
            struct arena_object* ao;
            uint nf;  /* ao->nfreepools */
            /* freeblock wasn't NULL, so the pool wasn't full,
             * and the pool is in a usedpools[] list.
             */
             /* 
              * pool没有full，且至少还有内存被分配出去: usable状态
              */
            if (--pool->ref.count != 0) {
                /* pool isn't empty:  leave it in usedpools */
                UNLOCK();
                return;
            }
            /* Pool is now empty:  unlink from usedpools, and
             * link to the front of freepools.  This ensures that
             * previously freed pools will be allocated later
             * (being not referenced, they are perhaps paged out).
             */
            /* pool的内存都回收了: empty状态
             * 将其从usedpools中去除
             */
            next = pool->nextpool;
            prev = pool->prevpool;
            next->prevpool = prev;
            prev->nextpool = next;
            /* Link the pool to freepools.  This is a singly-linked
             * list, and pool->prevpool isn't used there.
             */
            /* Pool放入到对应的arena->freepools中 */
            ao = &arenas[pool->arenaindex];
            pool->nextpool = ao->freepools;
            ao->freepools = pool;
            nf = ++ao->nfreepools;
            /* All the rest is arena management.  We just freed
             * a pool, and there are 4 cases for arena mgmt:
             * 1. If all the pools are free, return the arena to
             *    the system free().
             * 2. If this is the only free pool in the arena,
             *    add the arena back to the `usable_arenas` list.
             * 3. If the "next" arena has a smaller count of free
             *    pools, we have to "slide this arena right" to
             *    restore that usable_arenas is sorted in order of
             *    nfreepools.
             * 4. Else there's nothing more to do.
             */
            /* 1. 所有pool的内存都没有分配出去，则可以将1个arena的内存全部释放
             *    注意并不是释放arena_object，而是里面的address标记的256KB内存
             * 2. 只有1个空闲的Pool，则将arena放到usable_arenas链表的最后
             * 3. 重新排序usable_arenas，让空闲Pool多的排在后面不容易再被分配出去，
             *    使其有更大的可能整体被回收
             * 4. 无
             */
            if (nf == ao->ntotalpools) {
                /* Case 1.  First unlink ao from usable_arenas.
                 */
                assert(ao->prevarena == NULL ||
                       ao->prevarena->address != 0);
                assert(ao ->nextarena == NULL ||
                       ao->nextarena->address != 0);
                /* Fix the pointer in the prevarena, or the
                 * usable_arenas pointer.
                 */
                /* 
                 * 将arena从链表中拆出来 
                 */
                if (ao->prevarena == NULL) {
                    usable_arenas = ao->nextarena;
                    assert(usable_arenas == NULL ||
                           usable_arenas->address != 0);
                }
                else {
                    assert(ao->prevarena->nextarena == ao);
                    ao->prevarena->nextarena =
                        ao->nextarena;
                }
                /* Fix the pointer in the nextarena. */
                if (ao->nextarena != NULL) {
                    assert(ao->nextarena->prevarena == ao);
                    ao->nextarena->prevarena =
                        ao->prevarena;
                }
                /* Record that this arena_object slot is
                 * available to be reused.
                 */
                /* 放入unused_arena_objects链表中 */
                ao->nextarena = unused_arena_objects;
                unused_arena_objects = ao;
                /* Free the entire arena. */
                /* 释放其中的ARENA_SIZE内存 */
#ifdef ARENAS_USE_MMAP
                munmap((void *)ao->address, ARENA_SIZE);
#else
                free((void *)ao->address);
#endif
                ao->address = 0;                        /* mark unassociated */
                --narenas_currently_allocated;
                UNLOCK();
                return;
            }
            if (nf == 1) {
                /* Case 2.  Put ao at the head of
                 * usable_arenas.  Note that because
                 * ao->nfreepools was 0 before, ao isn't
                 * currently on the usable_arenas list.
                 */
                /* 当nf==1时，说明在释放刚才的Pool之前，整个arena的pools都被分配出去了，
                 * 所以之前肯定不在usabel_arenas中，于是放入usable_arenas表的最前面，即:
                 * 越满的越被容易再次分配出去
                 */
                ao->nextarena = usable_arenas;
                ao->prevarena = NULL;
                if (usable_arenas)
                    usable_arenas->prevarena = ao;
                usable_arenas = ao;
                assert(usable_arenas->address != 0);
                UNLOCK();
                return;
            }
            /* If this arena is now out of order, we need to keep
             * the list sorted.  The list is kept sorted so that
             * the "most full" arenas are used first, which allows
             * the nearly empty arenas to be completely freed.  In
             * a few un-scientific tests, it seems like this
             * approach allowed a lot more memory to be freed.
             */
            if (ao->nextarena == NULL ||
                         nf <= ao->nextarena->nfreepools) {
                /* Case 4.  Nothing to do. */
                UNLOCK();
                return;
            }
            /* Case 3:  We have to move the arena towards the end
             * of the list, because it has more free pools than
             * the arena to its right.
             * First unlink ao from usable_arenas.
             */
            if (ao->prevarena != NULL) {
                /* ao isn't at the head of the list */
                assert(ao->prevarena->nextarena == ao);
                ao->prevarena->nextarena = ao->nextarena;
            }
            else {
                /* ao is at the head of the list */
                assert(usable_arenas == ao);
                usable_arenas = ao->nextarena;
            }
            ao->nextarena->prevarena = ao->prevarena;
            /* Locate the new insertion point by iterating over
             * the list, using our nextarena pointer.
             */
            while (ao->nextarena != NULL &&
                            nf > ao->nextarena->nfreepools) {
                ao->prevarena = ao->nextarena;
                ao->nextarena = ao->nextarena->nextarena;
            }
            /* Insert ao at this point. */
            assert(ao->nextarena == NULL ||
                ao->prevarena == ao->nextarena->prevarena);
            assert(ao->prevarena->nextarena == ao->nextarena);
            ao->prevarena->nextarena = ao;
            if (ao->nextarena != NULL)
                ao->nextarena->prevarena = ao;
            /* Verify that the swaps worked. */
            assert(ao->nextarena == NULL ||
                      nf <= ao->nextarena->nfreepools);
            assert(ao->prevarena == NULL ||
                      nf > ao->prevarena->nfreepools);
            assert(ao->nextarena == NULL ||
                ao->nextarena->prevarena == ao);
            assert((usable_arenas == ao &&
                ao->prevarena == NULL) ||
                ao->prevarena->nextarena == ao);
            UNLOCK();
            return;
        }
        /* Pool was full, so doesn't currently live in any list:
         * link it to the front of the appropriate usedpools[] list.
         * This mimics LRU pool usage for new allocations and
         * targets optimal filling when several pools contain
         * blocks of the same size class.
         */
        /* Pool所在的arena没有被回收，Pool有部分内存被分配出去 */
        --pool->ref.count;
        assert(pool->ref.count > 0);            /* else the pool is empty */
        size = pool->szidx;
        next = usedpools[size + size];
        prev = next->prevpool;
        /* insert pool before next:   prev <-> pool <-> next */
        pool->nextpool = next;
        pool->prevpool = prev;
        next->prevpool = pool;
        prev->nextpool = pool;
        UNLOCK();
        return;
    }
#ifdef WITH_VALGRIND
redirect:
#endif
    /* We didn't allocate this address. */
    free(p);
}
/* realloc.  If p is NULL, this acts like malloc(nbytes).  Else if nbytes==0,
 * then as the Python docs promise, we do not treat this like free(p), and
 * return a non-NULL result.
 */
#undef PyObject_Realloc
ATTRIBUTE_NO_ADDRESS_SAFETY_ANALYSIS
void *
PyObject_Realloc(void *p, size_t nbytes)
{
    void *bp;
    poolp pool;
    size_t size;
#ifndef Py_USING_MEMORY_DEBUGGER
    uint arenaindex_temp;
#endif
    if (p == NULL)
        return PyObject_Malloc(nbytes);
    /*
     * Limit ourselves to PY_SSIZE_T_MAX bytes to prevent security holes.
     * Most python internals blindly use a signed Py_ssize_t to track
     * things without checking for overflows or negatives.
     * As size_t is unsigned, checking for nbytes < 0 is not required.
     */
    if (nbytes > PY_SSIZE_T_MAX)
        return NULL;
#ifdef WITH_VALGRIND
    /* Treat running_on_valgrind == -1 the same as 0 */
    if (UNLIKELY(running_on_valgrind > 0))
        goto redirect;
#endif
    pool = POOL_ADDR(p);
    if (Py_ADDRESS_IN_RANGE(p, pool)) {
        /* We're in charge of this block */
        size = INDEX2SIZE(pool->szidx);
        /* 收缩内存至少75%才实际操作 */
        if (nbytes <= size) {
            /* The block is staying the same or shrinking.  If
             * it's shrinking, there's a tradeoff:  it costs
             * cycles to copy the block to a smaller size class,
             * but it wastes memory not to copy it.  The
             * compromise here is to copy on shrink only if at
             * least 25% of size can be shaved off.
             */
            if (4 * nbytes > 3 * size) {
                /* It's the same,
                 * or shrinking and new/old > 3/4.
                 */
                return p;
            }
            size = nbytes;
        }
        bp = PyObject_Malloc(nbytes);
        if (bp != NULL) {
            memcpy(bp, p, size);
            PyObject_Free(p);
        }
        return bp;
    }
#ifdef WITH_VALGRIND
 redirect:
#endif
    /* We're not managing this block.  If nbytes <=
     * SMALL_REQUEST_THRESHOLD, it's tempting to try to take over this
     * block.  However, if we do, we need to copy the valid data from
     * the C-managed block to one of our blocks, and there's no portable
     * way to know how much of the memory space starting at p is valid.
     * As bug 1185883 pointed out the hard way, it's possible that the
     * C-managed block is "at the end" of allocated VM space, so that
     * a memory fault can occur if we try to copy nbytes bytes starting
     * at p.  Instead we punt:  let C continue to manage this block.
     */
    if (nbytes)
        return realloc(p, nbytes);
    /* C doesn't define the result of realloc(p, 0) (it may or may not
     * return NULL then), but Python's docs promise that nbytes==0 never
     * returns NULL.  We don't pass 0 to realloc(), to avoid that endcase
     * to begin with.  Even then, we can't be sure that realloc() won't
     * return NULL.
     */
    bp = realloc(p, 1);
    return bp ? bp : p;
}
#else   /* ! WITH_PYMALLOC */
/*==========================================================================*/
/* pymalloc not enabled:  Redirect the entry points to malloc.  These will
 * only be used by extensions that are compiled with pymalloc enabled. */
void *
PyObject_Malloc(size_t n)
{
    return PyMem_MALLOC(n);
}
void *
PyObject_Realloc(void *p, size_t n)
{
    return PyMem_REALLOC(p, n);
}
void
PyObject_Free(void *p)
{
    PyMem_FREE(p);
}
#endif /* WITH_PYMALLOC */
#ifdef PYMALLOC_DEBUG
/*==========================================================================*/
/* A x-platform debugging allocator.  This doesn't manage memory directly,
 * it wraps a real allocator, adding extra debugging info to the memory blocks.
 */
/* Special bytes broadcast into debug memory blocks at appropriate times.
 * Strings of these are unlikely to be valid addresses, floats, ints or
 * 7-bit ASCII.
 */
#undef CLEANBYTE
#undef DEADBYTE
#undef FORBIDDENBYTE
#define CLEANBYTE      0xCB    /* clean (newly allocated) memory */
#define DEADBYTE       0xDB    /* dead (newly freed) memory */
#define FORBIDDENBYTE  0xFB    /* untouchable bytes at each end of a block */
/* We tag each block with an API ID in order to tag API violations */
#define _PYMALLOC_MEM_ID 'm'   /* the PyMem_Malloc() API */
#define _PYMALLOC_OBJ_ID 'o'   /* The PyObject_Malloc() API */
static size_t serialno = 0;     /* incremented on each debug {m,re}alloc */
/* serialno is always incremented via calling this routine.  The point is
 * to supply a single place to set a breakpoint.
 */
static void
bumpserialno(void)
{
    ++serialno;
}
#define SST SIZEOF_SIZE_T
/* Read sizeof(size_t) bytes at p as a big-endian size_t. */
static size_t
read_size_t(const void *p)
{
    const uchar *q = (const uchar *)p;
    size_t result = *q++;
    int i;
    for (i = SST; --i > 0; ++q)
        result = (result << 8) | *q;
    return result;
}
/* Write n as a big-endian size_t, MSB at address p, LSB at
 * p + sizeof(size_t) - 1.
 */
static void
write_size_t(void *p, size_t n)
{
    uchar *q = (uchar *)p + SST - 1;
    int i;
    for (i = SST; --i >= 0; --q) {
        *q = (uchar)(n & 0xff);
        n >>= 8;
    }
}
#ifdef Py_DEBUG
/* Is target in the list?  The list is traversed via the nextpool pointers.
 * The list may be NULL-terminated, or circular.  Return 1 if target is in
 * list, else 0.
 */
static int
pool_is_in_list(const poolp target, poolp list)
{
    poolp origlist = list;
    assert(target != NULL);
    if (list == NULL)
        return 0;
    do {
        if (target == list)
            return 1;
	lst->nextpool;
    } while (list != NULL && list != origlist);
    return 0;
}
#else
#define pool_is_in_list(X, Y) 1
#endif  /* Py_DEBUG */
/* Let S = sizeof(size_t).  The debug malloc asks for 4*S extra bytes and
   fills them with useful stuff, here calling the underlying malloc's result p:
p[0: S]
    Number of bytes originally asked for.  This is a size_t, big-endian (easier
    to read in a memory dump).
p[S: 2*S]
    Copies of FORBIDDENBYTE.  Used to catch under- writes and reads.
p[2*S: 2*S+n]
    The requested memory, filled with copies of CLEANBYTE.
    Used to catch reference to uninitialized memory.
    &p[2*S] is returned.  Note that this is 8-byte aligned if pymalloc
    handled the request itself.
p[2*S+n: 2*S+n+S]
    Copies of FORBIDDENBYTE.  Used to catch over- writes and reads.
p[2*S+n+S: 2*S+n+2*S]
    A serial number, incremented by 1 on each call to _PyObject_DebugMalloc
    and _PyObject_DebugRealloc.
    This is a big-endian size_t.
    If "bad memory" is detected later, the serial number gives an
    excellent way to set a breakpoint on the next run, to capture the
    instant at which this block was passed out.
*/
/* debug replacements for the PyMem_* memory API */
void *
_PyMem_DebugMalloc(size_t nbytes)
{
    return _PyObject_DebugMallocApi(_PYMALLOC_MEM_ID, nbytes);
}
void *
_PyMem_DebugRealloc(void *p, size_t nbytes)
{
    return _PyObject_DebugReallocApi(_PYMALLOC_MEM_ID, p, nbytes);
}
void
_PyMem_DebugFree(void *p)
{
    _PyObject_DebugFreeApi(_PYMALLOC_MEM_ID, p);
}
/* debug replacements for the PyObject_* memory API */
void *
_PyObject_DebugMalloc(size_t nbytes)
{
    return _PyObject_DebugMallocApi(_PYMALLOC_OBJ_ID, nbytes);
}
void *
_PyObject_DebugRealloc(void *p, size_t nbytes)
{
    return _PyObject_DebugReallocApi(_PYMALLOC_OBJ_ID, p, nbytes);
}
void
_PyObject_DebugFree(void *p)
{
    _PyObject_DebugFreeApi(_PYMALLOC_OBJ_ID, p);
}
void
_PyObject_DebugCheckAddress(const void *p)
{
    _PyObject_DebugCheckAddressApi(_PYMALLOC_OBJ_ID, p);
}
/* generic debug memory api, with an "id" to identify the API in use */
void *
_PyObject_DebugMallocApi(char id, size_t nbytes)
{
    uchar *p;           /* base address of malloc'ed block */
    uchar *tail;        /* p + 2*SST + nbytes == pointer to tail pad bytes */
    size_t total;       /* nbytes + 4*SST */
    bumpserialno();
    total = nbytes + 4*SST;
    if (total < nbytes)
        /* overflow:  can't represent total as a size_t */
        return NULL;
    p = (uchar *)PyObject_Malloc(total);
    if (p == NULL)
        return NULL;
    /* at p, write size (SST bytes), id (1 byte), pad (SST-1 bytes) */
    write_size_t(p, nbytes);
    p[SST] = (uchar)id;
    memset(p + SST + 1 , FORBIDDENBYTE, SST-1);
    if (nbytes > 0)
        memset(p + 2*SST, CLEANBYTE, nbytes);
    /* at tail, write pad (SST bytes) and serialno (SST bytes) */
    tail = p + 2*SST + nbytes;
    memset(tail, FORBIDDENBYTE, SST);
    write_size_t(tail + SST, serialno);
    return p + 2*SST;
}
/* The debug free first checks the 2*SST bytes on each end for sanity (in
   particular, that the FORBIDDENBYTEs with the api ID are still intact).
   Then fills the original bytes with DEADBYTE.
   Then calls the underlying free.
*/
void
_PyObject_DebugFreeApi(char api, void *p)
{
    uchar *q = (uchar *)p - 2*SST;  /* address returned from malloc */
    size_t nbytes;
    if (p == NULL)
        return;
    _PyObject_DebugCheckAddressApi(api, p);
    nbytes = read_size_t(q);
    nbytes += 4*SST;
    if (nbytes > 0)
        memset(q, DEADBYTE, nbytes);
    PyObject_Free(q);
}
void *
_PyObject_DebugReallocApi(char api, void *p, size_t nbytes)
{
    uchar *q = (uchar *)p;
    uchar *tail;
    size_t total;       /* nbytes + 4*SST */
    size_t original_nbytes;
    int i;
    if (p == NULL)
        return _PyObject_DebugMallocApi(api, nbytes);
    _PyObject_DebugCheckAddressApi(api, p);
    bumpserialno();
    original_nbytes = read_size_t(q - 2*SST);
    total = nbytes + 4*SST;
    if (total < nbytes)
        /* overflow:  can't represent total as a size_t */
        return NULL;
    if (nbytes < original_nbytes) {
        /* shrinking:  mark old extra memory dead */
        memset(q + nbytes, DEADBYTE, original_nbytes - nbytes + 2*SST);
    }
    /* Resize and add decorations. We may get a new pointer here, in which
     * case we didn't get the chance to mark the old memory with DEADBYTE,
     * but we live with that.
     */
    q = (uchar *)PyObject_Realloc(q - 2*SST, total);
    if (q == NULL)
        return NULL;
    write_size_t(q, nbytes);
    assert(q[SST] == (uchar)api);
    for (i = 1; i < SST; ++i)
        assert(q[SST + i] == FORBIDDENBYTE);
    q += 2*SST;
    tail = q + nbytes;
    memset(tail, FORBIDDENBYTE, SST);
    write_size_t(tail + SST, serialno);
    if (nbytes > original_nbytes) {
        /* growing:  mark new extra memory clean */
        memset(q + original_nbytes, CLEANBYTE,
               nbytes - original_nbytes);
    }
    return q;
}
/* Check the forbidden bytes on both ends of the memory allocated for p.
 * If anything is wrong, print info to stderr via _PyObject_DebugDumpAddress,
 * and call Py_FatalError to kill the program.
 * The API id, is also checked.
 */
 void
_PyObject_DebugCheckAddressApi(char api, const void *p)
{
    const uchar *q = (const uchar *)p;
    char msgbuf[64];
    char *msg;
    size_t nbytes;
    const uchar *tail;
    int i;
    char id;
    if (p == NULL) {
        msg = "didn't expect a NULL pointer";
        goto error;
    }
    /* Check the API id */
    id = (char)q[-SST];
    if (id != api) {
        msg = msgbuf;
        snprintf(msg, sizeof(msgbuf), "bad ID: Allocated using API '%c', verified using API '%c'", id, api);
        msgbuf[sizeof(msgbuf)-1] = 0;
        goto error;
    }
    /* Check the stuff at the start of p first:  if there's underwrite
     * corruption, the number-of-bytes field may be nuts, and checking
     * the tail could lead to a segfault then.
     */
    for (i = SST-1; i >= 1; --i) {
        if (*(q-i) != FORBIDDENBYTE) {
            msg = "bad leading pad byte";
            goto error;
        }
    }
    nbytes = read_size_t(q - 2*SST);
    tail = q + nbytes;
    for (i = 0; i < SST; ++i) {
        if (tail[i] != FORBIDDENBYTE) {
            msg = "bad trailing pad byte";
            goto error;
        }
    }
    return;
error:
    _PyObject_DebugDumpAddress(p);
    Py_FatalError(msg);
}
/* Display info to stderr about the memory block at p. */
void
_PyObject_DebugDumpAddress(const void *p)
{
    const uchar *q = (const uchar *)p;
    const uchar *tail;
    size_t nbytes, serial;
    int i;
    int ok;
    char id;
    fprintf(stderr, "Debug memory block at address p=%p:", p);
    if (p == NULL) {
        fprintf(stderr, "\n");
        return;
    }
    id = (char)q[-SST];
    fprintf(stderr, " API '%c'\n", id);
    nbytes = read_size_t(q - 2*SST);
    fprintf(stderr, "    %" PY_FORMAT_SIZE_T "u bytes originally "
                    "requested\n", nbytes);
    /* In case this is nuts, check the leading pad bytes first. */
    fprintf(stderr, "    The %d pad bytes at p-%d are ", SST-1, SST-1);
    ok = 1;
    for (i = 1; i <= SST-1; ++i) {
        if (*(q-i) != FORBIDDENBYTE) {
            ok = 0;
            break;
        }
    }
    if (ok)
        fputs("FORBIDDENBYTE, as expected.\n", stderr);
    else {
        fprintf(stderr, "not all FORBIDDENBYTE (0x%02x):\n",
            FORBIDDENBYTE);
        for (i = SST-1; i >= 1; --i) {
            const uchar byte = *(q-i);
            fprintf(stderr, "        at p-%d: 0x%02x", i, byte);
            if (byte != FORBIDDENBYTE)
                fputs(" *** OUCH", stderr);
            fputc('\n', stderr);
        }
        fputs("    Because memory is corrupted at the start, the "
              "count of bytes requested\n"
              "       may be bogus, and checking the trailing pad "
              "bytes may segfault.\n", stderr);
    }
    tail = q + nbytes;
    fprintf(stderr, "    The %d pad bytes at tail=%p are ", SST, tail);
    ok = 1;
    for (i = 0; i < SST; ++i) {
        if (tail[i] != FORBIDDENBYTE) {
            ok = 0;
            break;
        }
    }
    if (ok)
        fputs("FORBIDDENBYTE, as expected.\n", stderr);
    else {
        fprintf(stderr, "not all FORBIDDENBYTE (0x%02x):\n",
                FORBIDDENBYTE);
        for (i = 0; i < SST; ++i) {
            const uchar byte = tail[i];
            fprintf(stderr, "        at tail+%d: 0x%02x",
                    i, byte);
            if (byte != FORBIDDENBYTE)
                fputs(" *** OUCH", stderr);
            fputc('\n', stderr);
        }
    }
    serial = read_size_t(tail + SST);
    fprintf(stderr, "    The block was made by call #%" PY_FORMAT_SIZE_T
                    "u to debug malloc/realloc.\n", serial);
    if (nbytes > 0) {
        i = 0;
        fputs("    Data at p:", stderr);
        /* print up to 8 bytes at the start */
        while (q < tail && i < 8) {
            fprintf(stderr, " %02x", *q);
            ++i;
            ++q;
        }
        /* and up to 8 at the end */
        if (q < tail) {
            if (tail - q > 8) {
                fputs(" ...", stderr);
                q = tail - 8;
            }
            while (q < tail) {
                fprintf(stderr, " %02x", *q);
                ++q;
            }
        }
        fputc('\n', stderr);
    }
}
static size_t
printone(const char* msg, size_t value)
{
    int i, k;
    char buf[100];
    size_t origvalue = value;
    fputs(msg, stderr);
    for (i = (int)strlen(msg); i < 35; ++i)
        fputc(' ', stderr);
    fputc('=', stderr);
    /* Write the value with commas. */
    i = 22;
    buf[i--] = '\0';
    buf[i--] = '\n';
    k = 3;
    do {
        size_t nextvalue = value / 10;
        unsigned int digit = (unsigned int)(value - nextvalue * 10);
        value = nextvalue;
        buf[i--] = (char)(digit + '0');
        --k;
        if (k == 0 && value && i >= 0) {
            k = 3;
            buf[i--] = ',';
        }
    } while (value && i >= 0);
    while (i >= 0)
        buf[i--] = ' ';
    fputs(buf, stderr);
    return origvalue;
}
/* Print summary info to stderr about the state of pymalloc's structures.
 * In Py_DEBUG mode, also perform some expensive internal consistency
 * checks.
 */
void
_PyObject_DebugMallocStats(void)
{
    uint i;
    const uint numclasses = SMALL_REQUEST_THRESHOLD >> ALIGNMENT_SHIFT;
    /* # of pools, allocated blocks, and free blocks per class index */
    size_t numpools[SMALL_REQUEST_THRESHOLD >> ALIGNMENT_SHIFT];
    size_t numblocks[SMALL_REQUEST_THRESHOLD >> ALIGNMENT_SHIFT];
    size_t numfreeblocks[SMALL_REQUEST_THRESHOLD >> ALIGNMENT_SHIFT];
    /* total # of allocated bytes in used and full pools */
    size_t allocated_bytes = 0;
    /* total # of available bytes in used pools */
    size_t available_bytes = 0;
    /* # of free pools + pools not yet carved out of current arena */
    uint numfreepools = 0;
    /* # of bytes for arena alignment padding */
    size_t arena_alignment = 0;
    /* # of bytes in used and full pools used for pool_headers */
    size_t pool_header_bytes = 0;
    /* # of bytes in used and full pools wasted due to quantization,
     * i.e. the necessarily leftover space at the ends of used and
     * full pools.
     */
    size_t quantization = 0;
    /* # of arenas actually allocated. */
    size_t narenas = 0;
    /* running total -- should equal narenas * ARENA_SIZE */
    size_t total;
    char buf[128];
    fprintf(stderr, "Small block threshold = %d, in %u size classes.\n",
            SMALL_REQUEST_THRESHOLD, numclasses);
    for (i = 0; i < numclasses; ++i)
        numpools[i] = numblocks[i] = numfreeblocks[i] = 0;
    /* Because full pools aren't linked to from anything, it's easiest
     * to march over all the arenas.  If we're lucky, most of the memory
     * will be living in full pools -- would be a shame to miss them.
     */
    for (i = 0; i < maxarenas; ++i) {
        uint j;
        uptr base = arenas[i].address;
        /* Skip arenas which are not allocated. */
        if (arenas[i].address == (uptr)NULL)
            continue;
        narenas += 1;
        numfreepools += arenas[i].nfreepools;
        /* round up to pool alignment */
        if (base & (uptr)POOL_SIZE_MASK) {
            arena_alignment += POOL_SIZE;
            base &= ~(uptr)POOL_SIZE_MASK;
            base += POOL_SIZE;
        }
        /* visit every pool in the arena */
        assert(base <= (uptr) arenas[i].pool_address);
        for (j = 0;
                    base < (uptr) arenas[i].pool_address;
                    ++j, base += POOL_SIZE) {
            poolp p = (poolp)base;
            const uint sz = p->szidx;
            uint freeblocks;
            if (p->ref.count == 0) {
                /* currently unused */
                assert(pool_is_in_list(p, arenas[i].freepools));
                continue;
            }
            ++numpools[sz];
            numblocks[sz] += p->ref.count;
            freeblocks = NUMBLOCKS(sz) - p->ref.count;
            numfreeblocks[sz] += freeblocks;
#ifdef Py_DEBUG
            if (freeblocks > 0)
                assert(pool_is_in_list(p, usedpools[sz + sz]));
#endif
        }
    }
    assert(narenas == narenas_currently_allocated);
    fputc('\n', stderr);
    fputs("class   size   num pools   blocks in use  avail blocks\n"
          "-----   ----   ---------   -------------  ------------\n",
          stderr);
    for (i = 0; i < numclasses; ++i) {
        size_t p = numpools[i];
        size_t b = numblocks[i];
        size_t f = numfreeblocks[i];
        uint size = INDEX2SIZE(i);
        if (p == 0) {
            assert(b == 0 && f == 0);
            continue;
        }
        fprintf(stderr, "%5u %6u "
                        "%11" PY_FORMAT_SIZE_T "u "
                        "%15" PY_FORMAT_SIZE_T "u "
                        "%13" PY_FORMAT_SIZE_T "u\n",
                i, size, p, b, f);
        allocated_bytes += b * size;
        available_bytes += f * size;
        pool_header_bytes += p * POOL_OVERHEAD;
        quantization += p * ((POOL_SIZE - POOL_OVERHEAD) % size);
    }
    fputc('\n', stderr);
    (void)printone("# times object malloc called", serialno);
    (void)printone("# arenas allocated total", ntimes_arena_allocated);
    (void)printone("# arenas reclaimed", ntimes_arena_allocated - narenas);
    (void)printone("# arenas highwater mark", narenas_highwater);
    (void)printone("# arenas allocated current", narenas);
    PyOS_snprintf(buf, sizeof(buf),
        "%" PY_FORMAT_SIZE_T "u arenas * %d bytes/arena",
        narenas, ARENA_SIZE);
    (void)printone(buf, narenas * ARENA_SIZE);
    fputc('\n', stderr);
    total = printone("# bytes in allocated blocks", allocated_bytes);
    total += printone("# bytes in available blocks", available_bytes);
    PyOS_snprintf(buf, sizeof(buf),
        "%u unused pools * %d bytes", numfreepools, POOL_SIZE);
    total += printone(buf, (size_t)numfreepools * POOL_SIZE);
    total += printone("# bytes lost to pool headers", pool_header_bytes);
    total += printone("# bytes lost to quantization", quantization);
    total += printone("# bytes lost to arena alignment", arena_alignment);
    (void)printone("Total", total);
}
#endif  /* PYMALLOC_DEBUG */
#ifdef Py_USING_MEMORY_DEBUGGER
/* Make this function last so gcc won't inline it since the definition is
 * after the reference.
 */
int
Py_ADDRESS_IN_RANGE(void *P, poolp pool)
{
    uint arenaindex_temp = pool->arenaindex;
    return arenaindex_temp < maxarenas &&
           (uptr)P - arenas[arenaindex_temp].address < (uptr)ARENA_SIZE &&
           arenas[arenaindex_temp].address != 0;
}
#endifist

requests的高级功能-认证机制与redirect机制

发表于 2016-08-02 | 分类于 python ， requests |

requests中的认证机制

用户可以通过Session.request接口传入auth参数指定用户名和密码。auth参数可以是(username, password)的数组；也可以是HTTPBaiscAuth类似的实例，只要支持调用即可。

1
2
3


:param auth: (optional) Auth tuple or callable to enable
    Basic/Digest/Custom HTTP Auth.

认证信息主要从多个方面来获取

通过auth参数指定
否则，从URI中获取。（http://username:passwd@www.sina.com）
否则，从Session.auth中获取。( 通过session.auth=(username,passwd)设置 )
否则，从.netrc中获取

.netrc

requests也支持.netrc，.netrc用于记录访问的认证信息，具体的语法可以参考这里，大致语法如下。

machine definitions

认证信息

machine ftp.freebsd.org
	login anonymous
	password edwin@mavetju.org
machine myownmachine
	login myusername
	password mypassword

macro definitions

定义ftp bash登录后的执行命令

macdef uploadtest
	cd /pub/tests
	bin
	put filename.tar.gz
	quit
macdef dailyupload
	cd /pub/tests
	bin
	put daily-$1.tar.gz
	quit

requests中的redirect机制

当访问www.sina.com时，会发现requests中缓存了两个地址www.sina.com与www.sina.com.cn，因为前一个地址会被重定向到后一个地址上。当我们用curl工具直接访问会发现，该地址返回了301 Moved Permanently以及Location: http://www.sina.com.cn。于是requests会自动对重定向地址再次发起请求。

$ curl -i -X GET 'http://www.sina.com'
  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
100   178  100   178    0     0    712      0 --:--:-- --:--:-- --:--:--   712HTTP/1.1 301 Moved Permanently
Server: nginx
Date: Tue, 02 Aug 2016 05:20:52 GMT
Content-Type: text/html
Location: http://www.sina.com.cn/       # 重定向后的地址
Expires: Tue, 02 Aug 2016 05:22:52 GMT
Cache-Control: max-age=120
Age: 96
Content-Length: 178
X-Cache: HIT from xd33-78.sina.com.cn
<html>
<head><title>301 Moved Permanently</title></head>      # status_code和Moved Permanently
<body bgcolor="white">
<center><h1>301 Moved Permanently</h1></center>
<hr><center>nginx</center>
</body>
</html>

重定向后的访问逻辑主要在SessionRedirectMixin中（具体的请求过程分析参见这里)

resp = requests.get('http://www.sina.com')
resp   # 返回重定向后的真实访问请求response
Out[10]: <Response [200]>
resp.url
Out[11]: u'http://www.sina.com.cn/'
resp.history
Out[12]: [<Response [301]>]   # 多个重定向的访问response
resp.history[0].url
Out[13]: u'http://www.sina.com/'  # 最初传入的地址。

class SessionRedirectMixin(object):
    def resolve_redirects(self, resp, req, stream=False, timeout=None,
                          verify=True, cert=None, proxies=None, **adapter_kwargs):
        """Receives a Response. Returns a generator of Responses."""
        i = 0
        hist = [] # keep track of history
        
        # is_redirect就是返回的code in [301, 302, 303, 307, 308] 且headers中有Location字段
        while resp.is_redirect:
            prepared_request = req.copy()
            if i > 0:
                # Update history and keep track of redirects.
                hist.append(resp)
                new_hist = list(hist)
                resp.history = new_hist
            try:
                resp.content  # Consume socket so it can be released
            except (ChunkedEncodingError, ContentDecodingError, RuntimeError):
                resp.raw.read(decode_content=False)
            if i >= self.max_redirects:    # 默认值为30
                raise TooManyRedirects('Exceeded %s redirects.' % self.max_redirects, response=resp)
            # Release the connection back into the pool.
            resp.close()
            url = resp.headers['location']   # Location地址
            # Handle redirection without scheme (see: RFC 1808 Section 4)
            if url.startswith('//'):
                parsed_rurl = urlparse(resp.url)
                url = '%s:%s' % (parsed_rurl.scheme, url)
            # The scheme should be lower case...
            parsed = urlparse(url)
            url = parsed.geturl()
            # Facilitate relative 'location' headers, as allowed by RFC 7231.
            # (e.g. '/path/to/resource' instead of 'http://domain.tld/path/to/resource')
            # Compliant with RFC3986, we percent encode the url.
            if not parsed.netloc:
                url = urljoin(resp.url, requote_uri(url))
            else:
                url = requote_uri(url)
            prepared_request.url = to_native_string(url)
            # Cache the url, unless it redirects to itself.
            if resp.is_permanent_redirect and req.url != prepared_request.url:
                self.redirect_cache[req.url] = prepared_request.url
            self.rebuild_method(prepared_request, resp)
            # https://github.com/kennethreitz/requests/issues/1084
            if resp.status_code not in (codes.temporary_redirect, codes.permanent_redirect):
                if 'Content-Length' in prepared_request.headers:
                    del prepared_request.headers['Content-Length']
                prepared_request.body = None
            headers = prepared_request.headers
            try:
                del headers['Cookie']
            except KeyError:
                pass
            # Extract any cookies sent on the response to the cookiejar
            # in the new request. Because we've mutated our copied prepared
            # request, use the old one that we haven't yet touched.
            extract_cookies_to_jar(prepared_request._cookies, req, resp.raw)
            prepared_request._cookies.update(self.cookies)
            prepared_request.prepare_cookies(prepared_request._cookies)
            # Rebuild auth and proxy information.
            proxies = self.rebuild_proxies(prepared_request, proxies)
            self.rebuild_auth(prepared_request, resp)
            # Override the original request.
            req = prepared_request
            #　重新发起send操作。
            resp = self.send(
                req,
                stream=stream,
                timeout=timeout,
                verify=verify,
                cert=cert,
                proxies=proxies,
                allow_redirects=False,
                **adapter_kwargs
            )
            extract_cookies_to_jar(self.cookies, prepared_request, resp.raw)
            i += 1
            yield resp

一般情况下只有指定的方法能够重定向

1 2	允许重定向： GET OPTIONS HEAD 不能重定向：POST PUT PATCH DELETE

根据重定向返回的状态码和访问方法，对重定向地址的访问需要修改访问方法

原访问方法   返回的状态码  状态码名字     新访问方法
GET/OPTIONS     303      see_other      GET
GET/OPTIONS     302      found          GET
POST            301      moved          GET

requests的高级功能-超时时间

发表于 2016-07-30 | 分类于 python ， requests |

requests的高级功能

如果简单的使用requests，会发现（requesets.get..)使用了默认参数的HTTPAdapter,因此所有由HTTPAdapter初始化参数指定的功能都没有办法使用，例如：重试、缓存池大小、缓存连接池大小、缓存池是否堵塞等。当然，因为requests.get方式只会发起一次HTTP请求，所以缓存相关的都没有设置的必要。

# requests.sessions.Session
class Session(SessionRedirectMixin):
    def __init__(self):
        ....
        self.adapters = OrderedDict()
        self.mount('https://', HTTPAdapter())
        self.mount('http://', HTTPAdapter())
        
# requests.adapters.HTTPAdapter
class HTTPAdapter(BaseAdapter):
    def __init__(self, pool_connections=DEFAULT_POOLSIZE,
                 pool_maxsize=DEFAULT_POOLSIZE, max_retries=DEFAULT_RETRIES,
                 pool_block=DEFAULT_POOLBLOCK):

设置超时时间

超时时间可以通过timeout参数指定,可以详细为(connect_timeout, read_timeout)。

"""
        :param timeout: (optional) How long to wait for the server to send
            data before giving up, as a float, or a :ref:`(connect timeout,
            read timeout) <timeouts>` tuple.
"""

通过流程图可以看到，传递的timeout参数一直进入到HTTPAdapter.send内。


# requests.adapters.HTTPAdapter
    def send(self, request, stream=False, timeout=None, verify=True, cert=None, proxies=None):
        ...
        if isinstance(timeout, tuple):
            try:
                connect, read = timeout
                # timout被实例化为
                timeout = TimeoutSauce(connect=connect, read=read)
            except ValueError as e:
                # this may raise a string formatting error.
                err = ("Invalid timeout {0}. Pass a (connect, read) "
                       "timeout tuple, or a single float to set "
                       "both timeouts to the same value".format(timeout))
                raise ValueError(err)
        else:
            timeout = TimeoutSauce(connect=timeout, read=timeout)
        try:
            if not chunked:
                resp = conn.urlopen(
                    method=request.method,
                    url=url,
                    body=request.body,
                    headers=request.headers,
                    redirect=False,
                    assert_same_host=False,
                    preload_content=False,
                    decode_content=False,
                    retries=self.max_retries,
                    timeout=timeout    # 传入HTTPConnectionPool.urlopen
                )
                
# requests.packages.urllib3.util.timeout.py
class Timeout(object):
   
    def __init__(self, total=None, connect=_Default, read=_Default):
        self._connect = self._validate_timeout(connect, 'connect')
        self._read = self._validate_timeout(read, 'read')
        self.total = self._validate_timeout(total, 'total')
        self._start_connect = None
    def get_connect_duration(self):
        """ Gets the time elapsed since the call to :meth:`start_connect`.
        :return: Elapsed time.
        :rtype: float
        :raises urllib3.exceptions.TimeoutStateError: if you attempt
            to get duration for a timer that hasn't been started.
        """
        if self._start_connect is None:
            raise TimeoutStateError("Can't get connect duration for timer "
                                    "that has not started.")
        return current_time() - self._start_connect
    @property
    def read_timeout(self):
        if (self.total is not None and
                self.total is not self.DEFAULT_TIMEOUT and
                self._read is not None and
                self._read is not self.DEFAULT_TIMEOUT):
            # In case the connect timeout has not yet been established.
            if self._start_connect is None:
                return self._read
            # 如果值指定了total，则read_timeout是链接后剩余的事件
            return max(0, min(self.total - self.get_connect_duration(),
                              self._read))
        elif self.total is not None and self.total is not self.DEFAULT_TIMEOUT:
            return max(0, self.total - self.get_connect_duration())
        else:
            return self._read

然后实例化后的timeout传递给HTTPConnectionPool,其中的connect_timeout设置为conn.timeout然后一直传递到socket中，通过socket.settimeout设置起效。需要注意socket是在设置参数之后再执行的bind->connect操作。


# lib/socket.py
_GLOBAL_DEFAULT_TIMEOUT = object()
def create_connection(address, timeout=_GLOBAL_DEFAULT_TIMEOUT,
                      source_address=None):
    host, port = address
    err = None
    for res in getaddrinfo(host, port, 0, SOCK_STREAM):
        af, socktype, proto, canonname, sa = res
        sock = None
        try:
            sock = socket(af, socktype, proto)
            if timeout is not _GLOBAL_DEFAULT_TIMEOUT:
                sock.settimeout(timeout)    
            if source_address:
                sock.bind(source_address)
            sock.connect(sa)
            return sock
        except error as _:
            err = _
            if sock is not None:
                sock.close()
    if err is not None:
        raise err
    else:
        raise error("getaddrinfo returns an empty list")

其中的read_timeout在HTTPConnectionPool中设置。通过代码可以看到socket.settimeout设置的是socket所有操作的超时时间，在不同的阶段调用该函数就设置了接下来操作的超时时间，settimeout -> bind -> connect -> settimeout -> read。


# requests.packages.urllib3.connectionpool.py
class HTTPConnectionPool
    def _make_request(...):
        ....
        read_timeout = timeout_obj.read_timeout
        # App Engine doesn't have a sock attr
        if getattr(conn, 'sock', None):
            # In Python 3 socket.py will catch EAGAIN and return None when you
            # try and read into the file pointer created by http.client, which
            # instead raises a BadStatusLine exception. Instead of catching
            # the exception and assuming all BadStatusLine exceptions are read
            # timeouts, check for a zero timeout before making the request.
            if read_timeout == 0:
                raise ReadTimeoutError(
                    self, url, "Read timed out. (read timeout=%s)" % read_timeout)
            if read_timeout is Timeout.DEFAULT_TIMEOUT:
                conn.sock.settimeout(socket.getdefaulttimeout())
            else:  # None or a value
                # 同样通过settimeout函数设置，不过此时已经connect完了，接下来就只有read操作
                conn.sock.settimeout(read_timeout)