|
1 | | -# Python虚拟机中的函数机制 |
| 1 | +# Python 虚拟机中的函数机制 |
| 2 | + |
| 3 | +一个函数意味着一个新的作用域(命名空间),当 Python 虚拟机执行一个函数时,首先创建一个 PyFrameObject,然后执行函数的字节码(PyCodeObject)。 |
| 4 | + |
| 5 | + |
| 6 | +PyFunctionObject 的定义: |
| 7 | + |
| 8 | +```C |
| 9 | +typedef struct { |
| 10 | + PyObject_HEAD |
| 11 | + PyObject *func_code; /* A code object */ |
| 12 | + PyObject *func_globals; /* A dictionary (other mappings won't do) */ |
| 13 | + PyObject *func_defaults; /* NULL or a tuple */ |
| 14 | + PyObject *func_closure; /* NULL or a tuple of cell objects */ |
| 15 | + PyObject *func_doc; /* The __doc__ attribute, can be anything */ |
| 16 | + PyObject *func_name; /* The __name__ attribute, a string object */ |
| 17 | + PyObject *func_dict; /* The __dict__ attribute, a dict or NULL */ |
| 18 | + PyObject *func_weakreflist; /* List of weak references */ |
| 19 | + PyObject *func_module; /* The __module__ attribute, can be anything */ |
| 20 | + |
| 21 | + /* Invariant: |
| 22 | + * func_closure contains the bindings for func_code->co_freevars, so |
| 23 | + * PyTuple_Size(func_closure) == PyCode_GetNumFree(func_code) |
| 24 | + * (func_closure may be NULL if PyCode_GetNumFree(func_code) == 0). |
| 25 | + */ |
| 26 | +} PyFunctionObject; |
| 27 | + |
| 28 | +``` |
| 29 | + |
| 30 | +PyCodeObject 是在编译时产生的,包含函数的一些静态信息,如 co_consts, co_names, co_code 等,而 PyFunctionObject 则是执行 def 语句时动态产生的,包含 PyCodeObject 以及其它动态的信息,如 func_globals。 |
| 31 | + |
| 32 | +一个函数只有一个 PyCodeObject,但是却会产生多个 PyFunctionObject,每次调用函数时都会创建一个新的 PyFunctionObject,该 PyFunctionObject 关联至唯一的 PyCodeObject。 |
| 33 | + |
| 34 | +## 函数调用流程 |
| 35 | + |
| 36 | +```python |
| 37 | +[func_0.py] |
| 38 | +def f(): |
| 39 | +0 LOAD_CONST 0 (code object f) |
| 40 | +3 MAKE_FUNCTION 0 |
| 41 | +6 STORE_NAME 0 (f) |
| 42 | + print "Function" |
| 43 | + 0 LOAD_CONST 1 (“Function”) |
| 44 | + 3 PRINT_ITEM |
| 45 | + 4 PRINT_NEWLINE |
| 46 | + 5 LOAD_CONST 0 (None) |
| 47 | + 8 RETURN_VALUE |
| 48 | + |
| 49 | +f() |
| 50 | +9 LOAD_NAME 0 (f) |
| 51 | +12 CALL_FUNCTION 0 |
| 52 | +15 POP_TOP |
| 53 | +16 LOAD_CONST 1 (None) |
| 54 | +19 RETURN_VALUE |
| 55 | + |
| 56 | +``` |
| 57 | + |
| 58 | +重点指令:MAKE_FUNCTION,CALL_FUNCTION。 |
| 59 | + |
| 60 | +```C |
| 61 | +case MAKE_FUNCTION: |
| 62 | + v = POP(); /* code object */ |
| 63 | + x = PyFunction_New(v, f->f_globals); |
| 64 | + Py_DECREF(v); |
| 65 | + /* XXX Maybe this should be a separate opcode? */ |
| 66 | + if (x != NULL && oparg> 0) { |
| 67 | + // oparg 表示 “具有默认值的参数” 的个数 |
| 68 | + v = PyTuple_New(oparg); |
| 69 | + if (v == NULL) { |
| 70 | + Py_DECREF(x); |
| 71 | + x = NULL; |
| 72 | + break; |
| 73 | + } |
| 74 | + while (--oparg>= 0) { |
| 75 | + w = POP(); |
| 76 | + PyTuple_SET_ITEM(v, oparg, w); |
| 77 | + } |
| 78 | + err = PyFunction_SetDefaults(x, v); |
| 79 | + Py_DECREF(v); |
| 80 | + } |
| 81 | + PUSH(x); |
| 82 | + break; |
| 83 | +``` |
| 84 | +
|
| 85 | +```C |
| 86 | +case CALL_FUNCTION: |
| 87 | + { |
| 88 | + PyObject **sp; |
| 89 | + PCALL(PCALL_ALL); |
| 90 | + sp = stack_pointer; |
| 91 | +#ifdef WITH_TSC |
| 92 | + x = call_function(&sp, oparg, &intr0, &intr1); |
| 93 | +#else |
| 94 | + x = call_function(&sp, oparg); |
| 95 | +#endif |
| 96 | + stack_pointer = sp; |
| 97 | + PUSH(x); // 函数返回值 |
| 98 | + if (x != NULL) |
| 99 | + continue; |
| 100 | + break; |
| 101 | + } |
| 102 | +
|
| 103 | +static PyObject * |
| 104 | +call_function(PyObject ***pp_stack, int oparg |
| 105 | +#ifdef WITH_TSC |
| 106 | + , uint64* pintr0, uint64* pintr1 |
| 107 | +#endif |
| 108 | + ) |
| 109 | +{ |
| 110 | + // oparg 是 short 类型,高字节表示 nk(实参中关键字参数个数),低字节表示 na(实参中位置参数个数) |
| 111 | + int na = oparg & 0xff; |
| 112 | + int nk = (oparg>>8) & 0xff; |
| 113 | + int n = na + 2 * nk; |
| 114 | + PyObject **pfunc = (*pp_stack) - n - 1; |
| 115 | + PyObject *func = *pfunc; |
| 116 | + PyObject *x, *w; |
| 117 | +
|
| 118 | + /* Always dispatch PyCFunction first, because these are |
| 119 | + presumed to be the most frequent callable object. |
| 120 | + */ |
| 121 | + if (PyCFunction_Check(func) && nk == 0) { |
| 122 | + int flags = PyCFunction_GET_FLAGS(func); |
| 123 | + PyThreadState *tstate = PyThreadState_GET(); |
| 124 | +
|
| 125 | + PCALL(PCALL_CFUNCTION); |
| 126 | + if (flags & (METH_NOARGS | METH_O)) { |
| 127 | + PyCFunction meth = PyCFunction_GET_FUNCTION(func); |
| 128 | + PyObject *self = PyCFunction_GET_SELF(func); |
| 129 | + if (flags & METH_NOARGS && na == 0) { |
| 130 | + C_TRACE(x, (*meth)(self,NULL)); |
| 131 | + } |
| 132 | + else if (flags & METH_O && na == 1) { |
| 133 | + PyObject *arg = EXT_POP(*pp_stack); |
| 134 | + C_TRACE(x, (*meth)(self,arg)); |
| 135 | + Py_DECREF(arg); |
| 136 | + } |
| 137 | + else { |
| 138 | + err_args(func, flags, na); |
| 139 | + x = NULL; |
| 140 | + } |
| 141 | + } |
| 142 | + else { |
| 143 | + PyObject *callargs; |
| 144 | + callargs = load_args(pp_stack, na); |
| 145 | + READ_TIMESTAMP(*pintr0); |
| 146 | + C_TRACE(x, PyCFunction_Call(func,callargs,NULL)); |
| 147 | + READ_TIMESTAMP(*pintr1); |
| 148 | + Py_XDECREF(callargs); |
| 149 | + } |
| 150 | + } else { |
| 151 | + if (PyMethod_Check(func) && PyMethod_GET_SELF(func) != NULL) { |
| 152 | + /* optimize access to bound methods */ |
| 153 | + PyObject *self = PyMethod_GET_SELF(func); |
| 154 | + PCALL(PCALL_METHOD); |
| 155 | + PCALL(PCALL_BOUND_METHOD); |
| 156 | + Py_INCREF(self); |
| 157 | + func = PyMethod_GET_FUNCTION(func); |
| 158 | + Py_INCREF(func); |
| 159 | + Py_DECREF(*pfunc); |
| 160 | + *pfunc = self; |
| 161 | + na++; |
| 162 | + n++; |
| 163 | + } else |
| 164 | + Py_INCREF(func); |
| 165 | + READ_TIMESTAMP(*pintr0); |
| 166 | + if (PyFunction_Check(func)) |
| 167 | + x = fast_function(func, pp_stack, n, na, nk); |
| 168 | + else |
| 169 | + x = do_call(func, pp_stack, na, nk); |
| 170 | + READ_TIMESTAMP(*pintr1); |
| 171 | + Py_DECREF(func); |
| 172 | + } |
| 173 | +
|
| 174 | + /* Clear the stack of the function object. Also removes |
| 175 | + the arguments in case they weren't consumed already |
| 176 | + (fast_function() and err_args() leave them on the stack). |
| 177 | + */ |
| 178 | + while ((*pp_stack) > pfunc) { |
| 179 | + w = EXT_POP(*pp_stack); |
| 180 | + Py_DECREF(w); |
| 181 | + PCALL(PCALL_POP); |
| 182 | + } |
| 183 | + return x; |
| 184 | +} |
| 185 | +``` |
| 186 | + |
| 187 | +`fast_function` 和 `PyEval_EvalCodeEx` 的代码太长,就不复制粘贴了。源代码(branch: ch11)有一些注释。 |
| 188 | + |
| 189 | + |
| 190 | +几个和函数参数有关的变量: |
| 191 | + |
| 192 | +- PyCodeObject->co_argcount |
| 193 | + |
| 194 | + 形参中参数的个数,不包括 *args 和 **kwargs |
| 195 | + |
| 196 | +- PyCodeObject->co_nlocals |
| 197 | + |
| 198 | + 函数局部变量的个数,包括参数个数(co_argcount + *args + **kwargs) |
| 199 | + |
| 200 | +- na |
| 201 | + |
| 202 | + 实参中位置参数的个数 |
| 203 | + |
| 204 | +- nw |
| 205 | + |
| 206 | + 实参中关键字参数的个数 |
| 207 | + |
| 208 | + |
| 209 | +f_localsplus 布局 |
| 210 | + |
| 211 | +``` |
| 212 | +PyFrameObject: |
| 213 | +
|
| 214 | ++------------+-------------------+---------------------------+ |
| 215 | +| | | | |
| 216 | +| frame_info | extras | valuestack | |
| 217 | +| | | | |
| 218 | ++--------------------------------+---------------------------+ |
| 219 | + | | |
| 220 | + <-------------+ f_localsplus +-----------------> |
| 221 | + | | |
| 222 | +
|
| 223 | +extras: |
| 224 | +
|
| 225 | ++----------+---------+-------+----------+----------+----------+ |
| 226 | +| | | | | | | |
| 227 | +| pos args | kw args | *args | **kwargs | cellvars | freevars | |
| 228 | +| | | | | | | |
| 229 | ++----------+---------+-------+----------+----------+----------+ |
| 230 | +
|
| 231 | +
|
| 232 | +``` |
| 233 | + |
| 234 | +cellvars 一般是外层函数使用,存放着被内层嵌套函数引用的变量 |
| 235 | + |
| 236 | +freevars 一般是内层嵌套函数使用,存放着对外层函数中变量的引用 |
| 237 | + |
| 238 | +## 常见的函数定义和调用 |
| 239 | + |
| 240 | +可以使用下面的例子去思考源码中 fast_function和 PyEval_EvalCodeEx的执行流程。 |
| 241 | + |
| 242 | +- 没有参数 |
| 243 | + |
| 244 | + ```py |
| 245 | + def f(): |
| 246 | + print 'hello world' |
| 247 | + |
| 248 | + f() |
| 249 | + |
| 250 | + # co_argcount = 0 |
| 251 | + # na = 0, nw = 0 |
| 252 | + ``` |
| 253 | + |
| 254 | + **符合 fast_function 条件** |
| 255 | + |
| 256 | +- 只有位置参数 |
| 257 | + |
| 258 | + ```py |
| 259 | + def f(a, b): |
| 260 | + print 'hello world' |
| 261 | + |
| 262 | + f(1, 2) |
| 263 | + # co_argcount = 2 |
| 264 | + # na = 2, nw = 0 |
| 265 | + |
| 266 | + f(a=1, b=2) |
| 267 | + # co_argcount = 2 |
| 268 | + # na = 0, nw = 2 |
| 269 | + ``` |
| 270 | + |
| 271 | + **f(1, 2) 符合 fast_function 条件** |
| 272 | + |
| 273 | + **f(a=1, b=2) 不符合 fast_function 条件** |
| 274 | + |
| 275 | +- 位置参数和关键字参数 |
| 276 | + |
| 277 | + ```py |
| 278 | + def f(a, b, c=1, d=2): |
| 279 | + print 'hello world' |
| 280 | + |
| 281 | + f(1, 2) |
| 282 | + |
| 283 | + f(1, 2, 3) |
| 284 | + |
| 285 | + f(1, 2, 3, 4) |
| 286 | + |
| 287 | + f(1, 2, c=3) |
| 288 | + |
| 289 | + f(1, 2, d=3) |
| 290 | + ``` |
| 291 | + |
| 292 | + |
| 293 | +- 可变参数 |
| 294 | + |
| 295 | + ```py |
| 296 | + def f(a, b, *args, **kwargs): |
| 297 | + print 'hello world' |
| 298 | + |
| 299 | + f(1, 2) |
| 300 | + |
| 301 | + f(1, 2, 3, 4, c=1, d=2) |
| 302 | + ``` |
0 commit comments