对python的LOAD_FAST/STORE_FAST感到困惑

当我写一些代码时，我发现一个有趣的事情：

def test():
  l = []
  for i in range(10):
    def f():pass
    print(f)
    #l.append(f)

test()

import dis
dis.dis(test)

输出是：

<function test.<locals>.f at 0x7f46c0b0d400>
<function test.<locals>.f at 0x7f46c0b0d488>
<function test.<locals>.f at 0x7f46c0b0d400>
<function test.<locals>.f at 0x7f46c0b0d488>
<function test.<locals>.f at 0x7f46c0b0d400>
<function test.<locals>.f at 0x7f46c0b0d488>
<function test.<locals>.f at 0x7f46c0b0d400>
<function test.<locals>.f at 0x7f46c0b0d488>
<function test.<locals>.f at 0x7f46c0b0d400>
<function test.<locals>.f at 0x7f46c0b0d488>
  6           0 BUILD_LIST               0
              3 STORE_FAST               0 (l)

  7           6 SETUP_LOOP              42 (to 51)
              9 LOAD_GLOBAL              0 (range)
             12 LOAD_CONST               1 (10)
             15 CALL_FUNCTION            1 (1 positional, 0 keyword pair)
             18 GET_ITER
        >>   19 FOR_ITER                28 (to 50)
             22 STORE_FAST               1 (i)

  8          25 LOAD_CONST               2 (<code object f at 0x7f46c0bd8420, file "ts.py", line 8>)
             28 LOAD_CONST               3 ('test.<locals>.f')
             31 MAKE_FUNCTION            0
             34 STORE_FAST               2 (f)

  9          37 LOAD_GLOBAL              1 (print)
             40 LOAD_FAST                2 (f)
             43 CALL_FUNCTION            1 (1 positional, 0 keyword pair)
             46 POP_TOP
             47 JUMP_ABSOLUTE           19
        >>   50 POP_BLOCK
        >>   51 LOAD_CONST               0 (None)
             54 RETURN_VALUE

when

def test():
  l = []
  for i in range(10):
    def f():pass
    print(f)
    l.append(f)

test()

import dis
dis.dis(test)

输出是：

<function test.<locals>.f at 0x7ff88ffe0400>
<function test.<locals>.f at 0x7ff88ffe0488>
<function test.<locals>.f at 0x7ff88ffe0510>
<function test.<locals>.f at 0x7ff88ffe0598>
<function test.<locals>.f at 0x7ff88ffe0620>
<function test.<locals>.f at 0x7ff88ffe06a8>
<function test.<locals>.f at 0x7ff88ffe0730>
<function test.<locals>.f at 0x7ff88ffe07b8>
<function test.<locals>.f at 0x7ff88ffe0840>
<function test.<locals>.f at 0x7ff88ffe08c8>
  6           0 BUILD_LIST               0
              3 STORE_FAST               0 (l)

  7           6 SETUP_LOOP              55 (to 64)
              9 LOAD_GLOBAL              0 (range)
             12 LOAD_CONST               1 (10)
             15 CALL_FUNCTION            1 (1 positional, 0 keyword pair)
             18 GET_ITER
        >>   19 FOR_ITER                41 (to 63)
             22 STORE_FAST               1 (i)

  8          25 LOAD_CONST               2 (<code object f at 0x7ff8900ab420, file "ts.py", line 8>)
             28 LOAD_CONST               3 ('test.<locals>.f')
             31 MAKE_FUNCTION            0
             34 STORE_FAST               2 (f)

  9          37 LOAD_GLOBAL              1 (print)
             40 LOAD_FAST                2 (f)
             43 CALL_FUNCTION            1 (1 positional, 0 keyword pair)
             46 POP_TOP

 10          47 LOAD_FAST                0 (l)
             50 LOAD_ATTR                2 (append)
             53 LOAD_FAST                2 (f)
             56 CALL_FUNCTION            1 (1 positional, 0 keyword pair)
             59 POP_TOP
             60 JUMP_ABSOLUTE           19
        >>   63 POP_BLOCK
        >>   64 LOAD_CONST               0 (None)
             67 RETURN_VALUE

If STORE_FAST“缓存”了f，为什么在第一个代码片段中，地址f是交替的？

在第二个片段中，它有两个LOAD_FAST，结果正常。

难道LOAD_FAST/STORE FAST做了一些未知的事情？

发生这种情况是因为在每次交替迭代中，在重新声明当前函数对象之后，旧函数对象会被重新声明。f没有留下任何引用，因此它被垃圾收集，Python 可以在下一次迭代中重新使用该内存空间。另一方面，在第二个列表中引用每个函数，因此它们永远不会被垃圾收集。

这是一个依赖于实现的事情，CPython 的垃圾收集是基于引用计数的。在 PyPy 上，输出有所不同：

$ ~/pypy-2.4.0-linux64/bin# ./pypy 
Python 2.7.8 (f5dcc2477b97, Sep 18 2014, 11:33:30)
[PyPy 2.4.0 with GCC 4.6.3] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>>> def test():
....     for i in range(10):
....         def f(): pass
....         print f
.... 
>>>> test()
<function f at 0x00007f055c77d5b0>
<function f at 0x00007f055c77d628>
<function f at 0x00007f055c77d6a0>
<function f at 0x00007f055c77d718>
<function f at 0x00007f055c77d790>
<function f at 0x00007f055c77d808>
<function f at 0x00007f055c77d880>
<function f at 0x00007f055c77d8f8>
<function f at 0x00007f055c77d970>
<function f at 0x00007f055c77d9e8>