关于 Python 循环引用

python解决循环引用的逻辑还是比较简单，今晚写了一个脚本层的查找unreachable object的方法，模拟的就是查找循环引用的算法

# -*- coding:utf-8 -*-

import gc
import sys

REACHABLE = -10001
TENTATIVELY_UNREACHABLE = -10002

def get_unreachable_objects():

    # 1. update refs
    gc_refs = {}
    for obj in gc.get_objects():
        assert id(obj) not in gc_refs
        # 在这个过程中会增加三个引用
        # 1. 放入list一个引用
        # 2. obj一个引用
        # 3. 调用getrefcount(obj)会产生一个临时引用
        gc_refs[id(obj)] = sys.getrefcount(obj) - 3  
    obj = None

    # 2. substract refs
    for obj in gc.get_objects():
        for child in gc.get_referents(obj):
            if id(child) in gc_refs:
                gc_refs[id(child)] -= 1
        child = None
    obj = None

    # 3. get unreachable
    is_running = True
    while is_running:
        for obj in gc.get_objects():
            assert id(obj) in gc_refs
            if gc_refs[id(obj)] == REACHABLE:
                continue
            if gc_refs[id(obj)] > 0:
                gc_refs[id(obj)] = REACHABLE
                for child in gc.get_referents(obj):
                    if id(child) in gc_refs:
                        if gc_refs[id(child)] == 0:
                            gc_refs[id(child)] = 1
                        elif gc_refs[id(child)] == TENTATIVELY_UNREACHABLE:
                            gc_refs[id(child)] = 1
                        else:
                            assert gc_refs[id(child)] > 0 or gc_refs[id(child)] == REACHABLE
                child = None
            elif gc_refs[id(obj)] == 0:
                gc_refs[id(obj)] = TENTATIVELY_UNREACHABLE
            else:
                assert gc_refs[id(obj)] in (REACHABLE, TENTATIVELY_UNREACHABLE)
        obj = None

        is_running = False
        for v in gc_refs.itervalues():
            if v not in (REACHABLE, TENTATIVELY_UNREACHABLE):
                is_running = True
                break
        v = None

    unreachable = [k for k, v in gc_refs.iteritems() if v == TENTATIVELY_UNREACHABLE]

    ret = []
    for obj in gc.get_objects():
        if id(obj) in unreachable:
            ret.append(obj)
    obj = None
    return ret

if __name__ == "__main__":
    class A(object): pass
    a = A()
    b = A()
    c = A()
    b.v = c
    c.v = b
    b_id = id(b)
    a_id = id(c)
    a = None
    b = None
    c = None
    print get_unreachable_objects()
    gc.set_debug(gc.DEBUG_LEAK | gc.DEBUG_STATS)
    gc.collect()

脚本分为三步：

• update_refs
• substract refs
• find unreachable objects

与Python GC的逻辑基本一样，目前在我们的开发服上测试还是有些问题