Python 字节码反汇编器#
参考：
先看给出函数示例：
def myfunc(alist):
return len(alist)
下面的命令可以用来显示 myfunc() 的反汇编：
import dis
dis.dis(myfunc)
1 0 RESUME 0
2 2 LOAD_GLOBAL 1 (NULL + len)
12 LOAD_FAST 0 (alist)
14 CALL 1
22 RETURN_VALUE
左上角的 2 是行号。
字节码分析#
字节码分析 API 允许将 Python 代码片段包装在 Bytecode 对象中，以便轻松访问已编译代码的详细信息。
bytecode = dis.Bytecode(myfunc)
for instr in bytecode:
print(instr.opname)
RESUME
LOAD_GLOBAL
LOAD_FAST
CALL
RETURN_VALUE
字节码#
使用三方库 bytecode。
安装：
pip install bytecode
抽象字节码#
下面使用抽象字节码执行 print('Hello World!')：
from bytecode import Instr, Bytecode
bytecode = Bytecode([Instr("LOAD_NAME", 'print'),
Instr("LOAD_CONST", 'Hello World!'),
Instr("CALL_FUNCTION", 1),
Instr("POP_TOP"),
Instr("LOAD_CONST", None),
Instr("RETURN_VALUE")])
code = bytecode.to_code()
exec(code)
---------------------------------------------------------------------------
KeyError Traceback (most recent call last)
File /opt/hostedtoolcache/Python/3.12.3/x64/lib/python3.12/site-packages/bytecode/instr.py:735, in BaseInstr._set(self, name, arg)
734 try:
--> 735 opcode = _opcode.opmap[name]
736 except KeyError:
KeyError: 'CALL_FUNCTION'
During handling of the above exception, another exception occurred:
ValueError Traceback (most recent call last)
Cell In[4], line 5
1 from bytecode import Instr, Bytecode
3 bytecode = Bytecode([Instr("LOAD_NAME", 'print'),
4 Instr("LOAD_CONST", 'Hello World!'),
----> 5 Instr("CALL_FUNCTION", 1),
6 Instr("POP_TOP"),
7 Instr("LOAD_CONST", None),
8 Instr("RETURN_VALUE")])
9 code = bytecode.to_code()
10 exec(code)
File /opt/hostedtoolcache/Python/3.12.3/x64/lib/python3.12/site-packages/bytecode/instr.py:536, in BaseInstr.__init__(self, name, arg, lineno, location)
528 def __init__(
529 self,
530 name: str,
(...)
534 location: Optional[InstrLocation] = None,
535 ) -> None:
--> 536 self._set(name, arg)
537 if location:
538 self._location = location
File /opt/hostedtoolcache/Python/3.12.3/x64/lib/python3.12/site-packages/bytecode/instr.py:737, in BaseInstr._set(self, name, arg)
735 opcode = _opcode.opmap[name]
736 except KeyError:
--> 737 raise ValueError(f"invalid operation name: {name}")
739 if opcode >= MIN_INSTRUMENTED_OPCODE:
740 raise ValueError(
741 f"operation {name} is an instrumented or pseudo opcode. "
742 "Only base opcodes are supported"
743 )
ValueError: invalid operation name: CALL_FUNCTION
具体字节码#
使用具体字节码执行 print('Hello World!') 的示例：
from bytecode import ConcreteInstr, ConcreteBytecode
bytecode = ConcreteBytecode()
bytecode.names = ['print']
bytecode.consts = ['Hello World!', None]
bytecode.extend([ConcreteInstr("LOAD_NAME", 0),
ConcreteInstr("LOAD_CONST", 0),
ConcreteInstr("CALL_FUNCTION", 1),
ConcreteInstr("POP_TOP"),
ConcreteInstr("LOAD_CONST", 1),
ConcreteInstr("RETURN_VALUE")])
code = bytecode.to_code()
exec(code)
---------------------------------------------------------------------------
KeyError Traceback (most recent call last)
File /opt/hostedtoolcache/Python/3.12.3/x64/lib/python3.12/site-packages/bytecode/instr.py:735, in BaseInstr._set(self, name, arg)
734 try:
--> 735 opcode = _opcode.opmap[name]
736 except KeyError:
KeyError: 'CALL_FUNCTION'
During handling of the above exception, another exception occurred:
ValueError Traceback (most recent call last)
Cell In[5], line 8
4 bytecode.names = ['print']
5 bytecode.consts = ['Hello World!', None]
6 bytecode.extend([ConcreteInstr("LOAD_NAME", 0),
7 ConcreteInstr("LOAD_CONST", 0),
----> 8 ConcreteInstr("CALL_FUNCTION", 1),
9 ConcreteInstr("POP_TOP"),
10 ConcreteInstr("LOAD_CONST", 1),
11 ConcreteInstr("RETURN_VALUE")])
12 code = bytecode.to_code()
13 exec(code)
File /opt/hostedtoolcache/Python/3.12.3/x64/lib/python3.12/site-packages/bytecode/concrete.py:97, in ConcreteInstr.__init__(self, name, arg, lineno, location, extended_args)
84 def __init__(
85 self,
86 name: str,
(...)
94 # Python to properly compute the size and avoid messing up the jump
95 # targets
96 self._extended_args = extended_args
---> 97 super().__init__(name, arg, lineno=lineno, location=location)
File /opt/hostedtoolcache/Python/3.12.3/x64/lib/python3.12/site-packages/bytecode/instr.py:536, in BaseInstr.__init__(self, name, arg, lineno, location)
528 def __init__(
529 self,
530 name: str,
(...)
534 location: Optional[InstrLocation] = None,
535 ) -> None:
--> 536 self._set(name, arg)
537 if location:
538 self._location = location
File /opt/hostedtoolcache/Python/3.12.3/x64/lib/python3.12/site-packages/bytecode/concrete.py:118, in ConcreteInstr._set(self, name, arg)
113 def _set(
114 self,
115 name: str,
116 arg: int,
117 ) -> None:
--> 118 super()._set(name, arg)
119 size = 2
120 if arg is not UNSET:
File /opt/hostedtoolcache/Python/3.12.3/x64/lib/python3.12/site-packages/bytecode/instr.py:737, in BaseInstr._set(self, name, arg)
735 opcode = _opcode.opmap[name]
736 except KeyError:
--> 737 raise ValueError(f"invalid operation name: {name}")
739 if opcode >= MIN_INSTRUMENTED_OPCODE:
740 raise ValueError(
741 f"operation {name} is an instrumented or pseudo opcode. "
742 "Only base opcodes are supported"
743 )
ValueError: invalid operation name: CALL_FUNCTION
设置编译器标志#
Bytecode、ConcreteBytecode 和 ControlFlowGraph 实例都有 flags 属性，它是 CompilerFlag 枚举的实例。可以像操作任何二进制标志一样操作该值。
设置 OPTIMIZED flags：
from bytecode import Bytecode, CompilerFlags
bytecode = Bytecode()
bytecode.flags |= CompilerFlags.OPTIMIZED
清除 OPTIMIZED flags：
from bytecode import Bytecode, CompilerFlags
bytecode = Bytecode()
bytecode.flags ^= CompilerFlags.OPTIMIZED
可以使用 update_flags 方法根据存储在代码对象中的指令更新 flags。
简单的循环#
for x in (1, 2, 3): print(x)：
from bytecode import Label, Instr, Bytecode
loop_start = Label()
loop_done = Label()
loop_exit = Label()
code = Bytecode(
[
# Python 3.8 removed SETUP_LOOP
Instr("LOAD_CONST", (1, 2, 3)),
Instr("GET_ITER"),
loop_start,
Instr("FOR_ITER", loop_exit),
Instr("STORE_NAME", "x"),
Instr("LOAD_NAME", "print"),
Instr("LOAD_NAME", "x"),
Instr("CALL_FUNCTION", 1),
Instr("POP_TOP"),
Instr("JUMP_ABSOLUTE", loop_start),
# Python 3.8 removed the need to manually manage blocks in loops
# This is now handled internally by the interpreter
loop_exit,
Instr("LOAD_CONST", None),
Instr("RETURN_VALUE"),
]
)
# The conversion to Python code object resolve jump targets:
# abstract labels are replaced with concrete offsets
code = code.to_code()
exec(code)
---------------------------------------------------------------------------
KeyError Traceback (most recent call last)
File /opt/hostedtoolcache/Python/3.12.3/x64/lib/python3.12/site-packages/bytecode/instr.py:735, in BaseInstr._set(self, name, arg)
734 try:
--> 735 opcode = _opcode.opmap[name]
736 except KeyError:
KeyError: 'CALL_FUNCTION'
During handling of the above exception, another exception occurred:
ValueError Traceback (most recent call last)
Cell In[8], line 16
4 loop_done = Label()
5 loop_exit = Label()
6 code = Bytecode(
7 [
8 # Python 3.8 removed SETUP_LOOP
9 Instr("LOAD_CONST", (1, 2, 3)),
10 Instr("GET_ITER"),
11 loop_start,
12 Instr("FOR_ITER", loop_exit),
13 Instr("STORE_NAME", "x"),
14 Instr("LOAD_NAME", "print"),
15 Instr("LOAD_NAME", "x"),
---> 16 Instr("CALL_FUNCTION", 1),
17 Instr("POP_TOP"),
18 Instr("JUMP_ABSOLUTE", loop_start),
19 # Python 3.8 removed the need to manually manage blocks in loops
20 # This is now handled internally by the interpreter
21 loop_exit,
22 Instr("LOAD_CONST", None),
23 Instr("RETURN_VALUE"),
24 ]
25 )
27 # The conversion to Python code object resolve jump targets:
28 # abstract labels are replaced with concrete offsets
29 code = code.to_code()
File /opt/hostedtoolcache/Python/3.12.3/x64/lib/python3.12/site-packages/bytecode/instr.py:536, in BaseInstr.__init__(self, name, arg, lineno, location)
528 def __init__(
529 self,
530 name: str,
(...)
534 location: Optional[InstrLocation] = None,
535 ) -> None:
--> 536 self._set(name, arg)
537 if location:
538 self._location = location
File /opt/hostedtoolcache/Python/3.12.3/x64/lib/python3.12/site-packages/bytecode/instr.py:737, in BaseInstr._set(self, name, arg)
735 opcode = _opcode.opmap[name]
736 except KeyError:
--> 737 raise ValueError(f"invalid operation name: {name}")
739 if opcode >= MIN_INSTRUMENTED_OPCODE:
740 raise ValueError(
741 f"operation {name} is an instrumented or pseudo opcode. "
742 "Only base opcodes are supported"
743 )
ValueError: invalid operation name: CALL_FUNCTION
条件调整#
print('yes' if test else 'no')：
from bytecode import Label, Instr, Bytecode
label_else = Label()
label_print = Label()
bytecode = Bytecode([Instr('LOAD_NAME', 'print'),
Instr('LOAD_NAME', 'test'),
Instr('POP_JUMP_IF_FALSE', label_else),
Instr('LOAD_CONST', 'yes'),
Instr('JUMP_FORWARD', label_print),
label_else,
Instr('LOAD_CONST', 'no'),
label_print,
Instr('CALL_FUNCTION', 1),
Instr('LOAD_CONST', None),
Instr('RETURN_VALUE')])
code = bytecode.to_code()
test = 0
exec(code)
test = 1
exec(code)
---------------------------------------------------------------------------
KeyError Traceback (most recent call last)
File /opt/hostedtoolcache/Python/3.12.3/x64/lib/python3.12/site-packages/bytecode/instr.py:735, in BaseInstr._set(self, name, arg)
734 try:
--> 735 opcode = _opcode.opmap[name]
736 except KeyError:
KeyError: 'CALL_FUNCTION'
During handling of the above exception, another exception occurred:
ValueError Traceback (most recent call last)
Cell In[9], line 13
3 label_else = Label()
4 label_print = Label()
5 bytecode = Bytecode([Instr('LOAD_NAME', 'print'),
6 Instr('LOAD_NAME', 'test'),
7 Instr('POP_JUMP_IF_FALSE', label_else),
8 Instr('LOAD_CONST', 'yes'),
9 Instr('JUMP_FORWARD', label_print),
10 label_else,
11 Instr('LOAD_CONST', 'no'),
12 label_print,
---> 13 Instr('CALL_FUNCTION', 1),
14 Instr('LOAD_CONST', None),
15 Instr('RETURN_VALUE')])
16 code = bytecode.to_code()
18 test = 0
File /opt/hostedtoolcache/Python/3.12.3/x64/lib/python3.12/site-packages/bytecode/instr.py:536, in BaseInstr.__init__(self, name, arg, lineno, location)
528 def __init__(
529 self,
530 name: str,
(...)
534 location: Optional[InstrLocation] = None,
535 ) -> None:
--> 536 self._set(name, arg)
537 if location:
538 self._location = location
File /opt/hostedtoolcache/Python/3.12.3/x64/lib/python3.12/site-packages/bytecode/instr.py:737, in BaseInstr._set(self, name, arg)
735 opcode = _opcode.opmap[name]
736 except KeyError:
--> 737 raise ValueError(f"invalid operation name: {name}")
739 if opcode >= MIN_INSTRUMENTED_OPCODE:
740 raise ValueError(
741 f"operation {name} is an instrumented or pseudo opcode. "
742 "Only base opcodes are supported"
743 )
ValueError: invalid operation name: CALL_FUNCTION
Control Flow Graph (CFG)#
为了分析或优化现有的代码，bytecode 提供了 ControlFlowGraph 类，它是控制流图(CFG)。
控制流图用于在转换为代码时进行堆栈深度分析。因为它比 CPython 更擅长识别死代码，所以它可以减少堆栈大小。
转储条件跳转示例的控制流图：
from bytecode import Label, Instr, Bytecode, ControlFlowGraph, dump_bytecode
label_else = Label()
label_print = Label()
bytecode = Bytecode([Instr('LOAD_NAME', 'print'),
Instr('LOAD_NAME', 'test'),
Instr('POP_JUMP_IF_FALSE', label_else),
Instr('LOAD_CONST', 'yes'),
Instr('JUMP_FORWARD', label_print),
label_else,
Instr('LOAD_CONST', 'no'),
label_print,
Instr('CALL_FUNCTION', 1),
Instr('LOAD_CONST', None),
Instr('RETURN_VALUE')])
blocks = ControlFlowGraph.from_bytecode(bytecode)
dump_bytecode(blocks)
---------------------------------------------------------------------------
KeyError Traceback (most recent call last)
File /opt/hostedtoolcache/Python/3.12.3/x64/lib/python3.12/site-packages/bytecode/instr.py:735, in BaseInstr._set(self, name, arg)
734 try:
--> 735 opcode = _opcode.opmap[name]
736 except KeyError:
KeyError: 'CALL_FUNCTION'
During handling of the above exception, another exception occurred:
ValueError Traceback (most recent call last)
Cell In[10], line 13
3 label_else = Label()
4 label_print = Label()
5 bytecode = Bytecode([Instr('LOAD_NAME', 'print'),
6 Instr('LOAD_NAME', 'test'),
7 Instr('POP_JUMP_IF_FALSE', label_else),
8 Instr('LOAD_CONST', 'yes'),
9 Instr('JUMP_FORWARD', label_print),
10 label_else,
11 Instr('LOAD_CONST', 'no'),
12 label_print,
---> 13 Instr('CALL_FUNCTION', 1),
14 Instr('LOAD_CONST', None),
15 Instr('RETURN_VALUE')])
17 blocks = ControlFlowGraph.from_bytecode(bytecode)
18 dump_bytecode(blocks)
File /opt/hostedtoolcache/Python/3.12.3/x64/lib/python3.12/site-packages/bytecode/instr.py:536, in BaseInstr.__init__(self, name, arg, lineno, location)
528 def __init__(
529 self,
530 name: str,
(...)
534 location: Optional[InstrLocation] = None,
535 ) -> None:
--> 536 self._set(name, arg)
537 if location:
538 self._location = location
File /opt/hostedtoolcache/Python/3.12.3/x64/lib/python3.12/site-packages/bytecode/instr.py:737, in BaseInstr._set(self, name, arg)
735 opcode = _opcode.opmap[name]
736 except KeyError:
--> 737 raise ValueError(f"invalid operation name: {name}")
739 if opcode >= MIN_INSTRUMENTED_OPCODE:
740 raise ValueError(
741 f"operation {name} is an instrumented or pseudo opcode. "
742 "Only base opcodes are supported"
743 )
ValueError: invalid operation name: CALL_FUNCTION
备注
block #1是开始块，以POP_JUMP_IF_FALSE条件跳转结束，跟着的是block #2block #2以JUMP_FORWARD无条件跳跃结束block #3不包含 jump，后面跟着block #4block #4是最终的块