想在 **字節碼級別** 追蹤 Python 函數執行：每條 opcode 執行前/後的堆棧狀態、幀信息、異常拋出瞬間。sys.settrace 粒度太粗（行級），決定動手擴展 CPython 3.5，實現一個 DEBUG_OP opcode + 運行時注入框架。
DEBUG_OP修改 Include/opcode.h：
#define DEBUG_OP 0 // 佔用原 NOP=0，或在末尾新增 #define POP_TOP 1 #define ROT_TWO 2 ...
修改 Python/opcode_targets.h 註冊跳轉表：
static void *opcode_targets[256] = {
&&TARGET_DEBUG_OP,
&&TARGET_POP_TOP,
...
};
TARGET(DEBUG_OP)（Python/ceval.c）TARGET(DEBUG_OP) {
PyObject *target = NULL, *value = NULL, *res = NULL, **sp = NULL, *tmp;
int i;
_Py_IDENTIFIER(op_target); // 固定回調函數名
target = _PyDict_GetItemId(f->f_globals, &PyId_op_target);
if (target == NULL && _PyErr_OCCURRED()) {
if (!PyErr_ExceptionMatches(PyExc_KeyError)) goto error;
PyErr_Clear(); DISPATCH(); // 無回調則靜默跳過
}
value = PyList_New(0); // 收集堆棧
for (i = 1; i <= STACK_LEVEL(); i++) {
tmp = PEEK(i);
if (tmp == NULL) tmp = Py_None;
PyList_Append(value, tmp);
}
Py_INCREF(target);
PUSH(target); PUSH(value); Py_INCREF(f); PUSH(f); // 參數：回調、堆棧列表、幀對象
sp = stack_pointer;
res = call_function(&sp, 2); // 調用 op_target(stack, frame)
stack_pointer = sp;
if (res == NULL) goto error;
Py_DECREF(res);
DISPATCH();
}
編譯：./configure && make 生成自定義 python 二進制。
核心難點：code.co_code 不可變、**分支指令偏移需同步調整**。
class MutableCodeObject:
attrs = ("co_argcount","co_kwonlyargcount","co_nlocals","co_stacksize",
"co_flags","co_code","co_consts","co_names","co_varnames",
"co_filename","co_name","co_firstlineno","co_lnotab",
"co_freevars","co_cellvars")
def __init__(self, code):
self.code = code
for a in self.attrs:
v = getattr(code, a)
setattr(self, a, list(v) if isinstance(v, tuple) else v)
def get_code(self):
args = [tuple(getattr(self,a)) if isinstance(getattr(self,a),list) else getattr(self,a) for a in self.attrs]
return self.code.__class__(*args)
JUMP_REL = {'FOR_ITER','JUMP_FORWARD','SETUP_LOOP','SETUP_WITH','SETUP_EXCEPT','SETUP_FINALLY'}
JUMP_ABS = {'POP_JUMP_IF_TRUE','POP_JUMP_IF_FALSE','JUMP_ABSOLUTE'}
def insert_debug_op(code, offset):
"""在指定字節偏移插入 DEBUG_OP(0x00)，同步修正後續跳轉目標"""
new_bytes = bytearray()
inserted = False
for instr in dis.Bytecode(code):
if instr.offset == offset:
new_bytes.append(0x00) # DEBUG_OP
inserted = True
# 相對跳轉：目標在插入點之後 → arg+1
if instr.opname in JUMP_REL and not inserted:
if offset < instr.offset + 3 + instr.arg:
new_bytes.extend(struct.pack(" 插入點 → arg+1
if instr.opname in JUMP_ABS:
if instr.arg > offset:
new_bytes.extend(struct.pack("
全函數注入
def add_debug_op_everywhere(code):
offsets = [i.offset for i in dis.Bytecode(code)]
for n, off in enumerate(sorted(offsets)):
code = insert_debug_op(code, off + n) # 每插一個，後續偏移+1
# 需同步增加 co_stacksize（DEBUG_OP 壓入 3 個參數）
m = MutableCodeObject(code)
m.co_stacksize += 3
return m.get_code()
第四步：Python 級追蹤器封裝
def op_target(stack, frame, exc=None):
if op_target.callback:
op_target.callback(stack, frame, exc)
class Trace:
def __init__(self, func):
self.func = func
def __call__(self, *a, **kw):
self._instrument(self.func)
op_target.callback = self.callback
try: return self.func(*a, **kw)
finally: op_target.callback = None
def _instrument(self, f):
if not hasattr(f, "op_debug") and hasattr(f, "__code__"):
f.__code__ = add_debug_op_everywhere(f.__code__)
f.__globals__["op_target"] = op_target
f.op_debug = True
def callback(self, stack, frame, exc): pass # 子類實現
# 示例：自動追蹤被調用函數
def do_auto_follow(self, stack, frame):
next_instr = FrameAnalyser(frame).next_instr()
if "CALL" in next_instr.opname:
f_idx = (next_instr.arg & 0xFF) + 2 * (next_instr.arg >> 8)
called = stack[f_idx]
if not hasattr(called, "op_debug"):
self._instrument(called)
效果演示
GitHub Repo（原文末尾鏈接）。
