PyCodeObject:代码对象,就是一段代码编译后形成的对象,函数中对应的就是函数体的代码编译结果。
PyFunctionObject :函数对象,它是对PyCodeObject的封装,相当于 PyCodeObject + 函数def定义这一行代码。它在PyCodeObject基础上增加了函数的名称、所属的模块、参数默认值、globals、builtins。
PyFrameObject:函数执行时对应的栈帧,它用于承载PyFunctionObject在执行时所需要的动态信息。包括函数的实参、函数执行时所需的栈、全局变量、局部变量、当前执行到的指令的编号。
以如下代码为例:
def foo(x, y=1): z = x + y return z*2 foo(2, 3)
其编译后的字节码为
1 0 LOAD_CONST 6 ((1,)) 2 LOAD_CONST 1 (<code object foo at 0x000002511F5B4F50, file "<dis>", line 1>) 4 LOAD_CONST 2 ('foo') 6 MAKE_FUNCTION 1 (defaults) 8 STORE_NAME 0 (foo) 5 10 LOAD_NAME 0 (foo) 12 LOAD_CONST 3 (2) 14 LOAD_CONST 4 (3) 16 CALL_FUNCTION 2 18 POP_TOP 20 LOAD_CONST 5 (None) 22 RETURN_VALUE Disassembly of <code object foo at 0x000002511F5B4F50, file "<dis>", line 1>: 2 0 LOAD_FAST 0 (x) 2 LOAD_FAST 1 (y) 4 BINARY_ADD 6 STORE_FAST 2 (z) 3 8 LOAD_FAST 2 (z) 10 LOAD_CONST 1 (2) 12 BINARY_MULTIPLY 14 RETURN_VALUE
PyCodeObject、PyFunctionObject、PyFrameObject三者的关系如下:
PyCodeObject在编译时确定,PyFunctionObject和PyFrameObject都在运行时生成。
其中PyFunctionObject在执行到函数定义指令MAKE_FUNCTION时生成,生成后是静态不变的。也就是说,一个函数一旦定义,其函数名参数默认值、函数绑定的globals和builtins信息不再变化。
PyFrameObject是动态可变的,其包含两层含义:
1)对同一个函数的每一次函数,都会生成一个新的PyFrameObject;
2)每个PyFrameObject在其生命周期内也是不断发生变化的,PyFrameObject承载着函数执行时所需要的所有动态信息。
MAKE_FUNCTION创建PyFunctionObject的过程:
// 创建PyFunctionObject PyObject * PyFunction_NewWithQualName(PyObject *code, PyObject *globals, PyObject *qualname) { // 获取解释器状态 PyThreadState *tstate = _PyThreadState_GET(); // 获取PyCodeObject PyCodeObject *code_obj = (PyCodeObject *)code; // 获取name和const PyObject *name = code_obj->co_name; if (!qualname) { qualname = name; } PyObject *consts = code_obj->co_consts; // 获取函数所属的__module__ // __module__: Use globals['__name__'] if it exists, or NULL. PyObject *module = _PyDict_GetItemIdWithError(globals, &PyId___name__); // 优先从globals中获取builtins,找不到则从当前栈帧中找,还找不到则以Python解释器中初始化定义的builtins为准 PyObject *builtins = NULL; builtins = _PyEval_BuiltinsFromGlobals(tstate, globals); // borrowed ref // 创建函数对象 PyFunctionObject *op = PyObject_GC_New(PyFunctionObject, &PyFunction_Type); // 设置参数 op->func_globals = globals; op->func_builtins = builtins; op->func_name = name; op->func_qualname = qualname; op->func_code = (PyObject*)code_obj; op->func_defaults = NULL; // No default positional arguments op->func_kwdefaults = NULL; // No default keyword arguments op->func_closure = NULL; op->func_dict = NULL; op->func_weakreflist = NULL; op->func_module = module; op->func_annotations = NULL; // Python函数调用的实现对应的C函数 op->vectorcall = _PyFunction_Vectorcall; return (PyObject *)op; }
CALL_FUNCTION执行函数过程:
// python函数调用 PyObject * _PyEval_Vector(PyThreadState *tstate, PyFrameConstructor *con, PyObject *locals, PyObject* const* args, size_t argcount, PyObject *kwnames) { // 创建栈帧 PyFrameObject *f = _PyEval_MakeFrameVector( tstate, con, locals, args, argcount, kwnames); // 执行栈帧 PyObject *retval = _PyEval_EvalFrame(tstate, f, 0); return retval; } static inline PyObject* _PyEval_EvalFrame(PyThreadState *tstate, PyFrameObject *f, int throwflag) { return tstate->interp->eval_frame(tstate, f, throwflag); } // 执行栈帧的实现 // 解释的eval_frame在解释器初始化时就设置成了_PyEval_EvalFrameDefault PyObject* _Py_HOT_FUNCTION _PyEval_EvalFrameDefault(PyThreadState *tstate, PyFrameObject *f, int throwflag) { // 连接到前一帧 CFrame *prev_cframe = tstate->cframe; // 切换当前帧 tstate->frame = f; PyCodeObject *co = f->f_code; first_instr = (_Py_CODEUNIT *) PyBytes_AS_STRING(co->co_code); next_instr = first_instr + f->f_lasti + 1; for (;;) { _Py_CODEUNIT word = *next_instr; opcode = _Py_OPCODE(word); oparg = _Py_OPARG(word); next_instr++; switch (opcode) { // 执行字节码 } } }