|
1 | 1 | # Python 中的 List 对象 |
2 | 2 |
|
| 3 | +PyListObject 定义如下: |
| 4 | + |
| 5 | +```C |
| 6 | +typedef struct { |
| 7 | + PyObject_VAR_HEAD |
| 8 | + /* Vector of pointers to list elements. list[0] is ob_item[0], etc. */ |
| 9 | + PyObject **ob_item; |
| 10 | + |
| 11 | + /* ob_item contains space for 'allocated' elements. The number |
| 12 | + * currently in use is ob_size. |
| 13 | + * Invariants: |
| 14 | + * 0 <= ob_size <= allocated |
| 15 | + * len(list) == ob_size |
| 16 | + * ob_item == NULL implies ob_size == allocated == 0 |
| 17 | + * list.sort() temporarily sets allocated to -1 to detect mutations. |
| 18 | + * |
| 19 | + * Items must normally not be NULL, except during construction when |
| 20 | + * the list is not yet visible outside the function that builds it. |
| 21 | + */ |
| 22 | + Py_ssize_t allocated; |
| 23 | +} PyListObject; |
| 24 | +``` |
| 25 | +代码注释已经说得足够明白了。 |
| 26 | + |
| 27 | +## 创建 PyListObject |
| 28 | + |
| 29 | +```C |
| 30 | +PyObject * |
| 31 | +PyList_New(Py_ssize_t size) |
| 32 | +{ |
| 33 | + PyListObject *op; |
| 34 | + size_t nbytes; |
| 35 | + |
| 36 | + if (size < 0) { |
| 37 | + PyErr_BadInternalCall(); |
| 38 | + return NULL; |
| 39 | + } |
| 40 | + nbytes = size * sizeof(PyObject *); |
| 41 | + /* Check for overflow */ |
| 42 | + if (nbytes / sizeof(PyObject *) != (size_t)size) |
| 43 | + return PyErr_NoMemory(); |
| 44 | + if (num_free_lists) { |
| 45 | + num_free_lists--; |
| 46 | + op = free_lists[num_free_lists]; |
| 47 | + _Py_NewReference((PyObject *)op); |
| 48 | + } else { |
| 49 | + op = PyObject_GC_New(PyListObject, &PyList_Type); |
| 50 | + if (op == NULL) |
| 51 | + return NULL; |
| 52 | + } |
| 53 | + if (size <= 0) |
| 54 | + op->ob_item = NULL; |
| 55 | + else { |
| 56 | + op->ob_item = (PyObject **) PyMem_MALLOC(nbytes); |
| 57 | + if (op->ob_item == NULL) { |
| 58 | + Py_DECREF(op); |
| 59 | + return PyErr_NoMemory(); |
| 60 | + } |
| 61 | + memset(op->ob_item, 0, nbytes); |
| 62 | + } |
| 63 | + op->ob_size = size; |
| 64 | + op->allocated = size; |
| 65 | + _PyObject_GC_TRACK(op); |
| 66 | + return (PyObject *) op; |
| 67 | +} |
| 68 | +``` |
| 69 | + |
| 70 | +## 设置元素 |
| 71 | + |
| 72 | +```C |
| 73 | +int |
| 74 | +PyList_SetItem(register PyObject *op, register Py_ssize_t i, |
| 75 | + register PyObject *newitem) |
| 76 | +{ |
| 77 | + register PyObject *olditem; |
| 78 | + register PyObject **p; |
| 79 | + if (!PyList_Check(op)) { |
| 80 | + Py_XDECREF(newitem); |
| 81 | + PyErr_BadInternalCall(); |
| 82 | + return -1; |
| 83 | + } |
| 84 | + if (i < 0 || i>= ((PyListObject *)op) -> ob_size) { |
| 85 | + Py_XDECREF(newitem); |
| 86 | + PyErr_SetString(PyExc_IndexError, |
| 87 | + "list assignment index out of range"); |
| 88 | + return -1; |
| 89 | + } |
| 90 | + p = ((PyListObject *)op) -> ob_item + i; |
| 91 | + olditem = *p; |
| 92 | + *p = newitem; |
| 93 | + Py_XDECREF(olditem); |
| 94 | + return 0; |
| 95 | +} |
| 96 | +``` |
| 97 | +** 旧元素的 refcnt - 1** |
| 98 | + |
| 99 | +## 插入新元素 |
| 100 | + |
| 101 | +```C |
| 102 | +static int |
| 103 | +ins1(PyListObject *self, Py_ssize_t where, PyObject *v) |
| 104 | +{ |
| 105 | + Py_ssize_t i, n = self->ob_size; |
| 106 | + PyObject **items; |
| 107 | + if (v == NULL) { |
| 108 | + PyErr_BadInternalCall(); |
| 109 | + return -1; |
| 110 | + } |
| 111 | + if (n == PY_SSIZE_T_MAX) { |
| 112 | + PyErr_SetString(PyExc_OverflowError, |
| 113 | + "cannot add more objects to list"); |
| 114 | + return -1; |
| 115 | + } |
| 116 | + |
| 117 | + if (list_resize(self, n+1) == -1) |
| 118 | + return -1; |
| 119 | + |
| 120 | + if (where < 0) { |
| 121 | + where += n; |
| 122 | + if (where < 0) |
| 123 | + where = 0; |
| 124 | + } |
| 125 | + if (where> n) |
| 126 | + where = n; |
| 127 | + items = self->ob_item; |
| 128 | + for (i = n; --i>= where; ) |
| 129 | + items[i+1] = items[i]; |
| 130 | + Py_INCREF(v); |
| 131 | + items[where] = v; |
| 132 | + return 0; |
| 133 | +} |
| 134 | + |
| 135 | +int |
| 136 | +PyList_Insert(PyObject *op, Py_ssize_t where, PyObject *newitem) |
| 137 | +{ |
| 138 | + if (!PyList_Check(op)) { |
| 139 | + PyErr_BadInternalCall(); |
| 140 | + return -1; |
| 141 | + } |
| 142 | + return ins1((PyListObject *)op, where, newitem); |
| 143 | +} |
| 144 | +``` |
| 145 | + |
| 146 | +插入元素的流程: |
| 147 | + |
| 148 | +- 调整 PyListObject 的尺寸,以便容下新元素。 |
| 149 | + |
| 150 | +- 从最后一个元素开始,每个元素往后以一步,直至 where(新元素插入的位置) |
| 151 | + |
| 152 | +- 将新元素放到 where。 |
| 153 | + |
| 154 | +list_resize 代码如下: |
| 155 | + |
| 156 | +```C |
| 157 | +/* Ensure ob_item has room for at least newsize elements, and set |
| 158 | + * ob_size to newsize. If newsize > ob_size on entry, the content |
| 159 | + * of the new slots at exit is undefined heap trash; it's the caller's |
| 160 | + * responsiblity to overwrite them with sane values. |
| 161 | + * The number of allocated elements may grow, shrink, or stay the same. |
| 162 | + * Failure is impossible if newsize <= self.allocated on entry, although |
| 163 | + * that partly relies on an assumption that the system realloc() never |
| 164 | + * fails when passed a number of bytes <= the number of bytes last |
| 165 | + * allocated (the C standard doesn't guarantee this, but it's hard to |
| 166 | + * imagine a realloc implementation where it wouldn't be true). |
| 167 | + * Note that self->ob_item may change, and even if newsize is less |
| 168 | + * than ob_size on entry. |
| 169 | + */ |
| 170 | +static int |
| 171 | +list_resize(PyListObject *self, Py_ssize_t newsize) |
| 172 | +{ |
| 173 | + PyObject **items; |
| 174 | + size_t new_allocated; |
| 175 | + Py_ssize_t allocated = self->allocated; |
| 176 | + |
| 177 | + /* Bypass realloc() when a previous overallocation is large enough |
| 178 | + to accommodate the newsize. If the newsize falls lower than half |
| 179 | + the allocated size, then proceed with the realloc() to shrink the list. |
| 180 | + */ |
| 181 | + if (allocated>= newsize && newsize >= (allocated >> 1)) { |
| 182 | + assert(self->ob_item != NULL || newsize == 0); |
| 183 | + self->ob_size = newsize; |
| 184 | + return 0; |
| 185 | + } |
| 186 | + |
| 187 | + /* This over-allocates proportional to the list size, making room |
| 188 | + * for additional growth. The over-allocation is mild, but is |
| 189 | + * enough to give linear-time amortized behavior over a long |
| 190 | + * sequence of appends() in the presence of a poorly-performing |
| 191 | + * system realloc(). |
| 192 | + * The growth pattern is: 0, 4, 8, 16, 25, 35, 46, 58, 72, 88, ... |
| 193 | + */ |
| 194 | + new_allocated = (newsize>> 3) + (newsize < 9 ? 3 : 6) + newsize; |
| 195 | + if (newsize == 0) |
| 196 | + new_allocated = 0; |
| 197 | + items = self->ob_item; |
| 198 | + if (new_allocated <= ((~(size_t)0) / sizeof(PyObject *))) |
| 199 | + PyMem_RESIZE(items, PyObject *, new_allocated); |
| 200 | + else |
| 201 | + items = NULL; |
| 202 | + if (items == NULL) { |
| 203 | + PyErr_NoMemory(); |
| 204 | + return -1; |
| 205 | + } |
| 206 | + self->ob_item = items; |
| 207 | + self->ob_size = newsize; |
| 208 | + self->allocated = new_allocated; |
| 209 | + return 0; |
| 210 | +} |
| 211 | +``` |
| 212 | + |
| 213 | +从代码中可以看出,为了避免频繁的调用 realloc,Python 会多分配一些空间。 |
| 214 | + |
| 215 | +## 添加元素 |
| 216 | + |
| 217 | +```C |
| 218 | +static int |
| 219 | +app1(PyListObject *self, PyObject *v) |
| 220 | +{ |
| 221 | + Py_ssize_t n = PyList_GET_SIZE(self); |
| 222 | + |
| 223 | + assert (v != NULL); |
| 224 | + if (n == PY_SSIZE_T_MAX) { |
| 225 | + PyErr_SetString(PyExc_OverflowError, |
| 226 | + "cannot add more objects to list"); |
| 227 | + return -1; |
| 228 | + } |
| 229 | + |
| 230 | + if (list_resize(self, n+1) == -1) |
| 231 | + return -1; |
| 232 | + |
| 233 | + Py_INCREF(v); |
| 234 | + PyList_SET_ITEM(self, n, v); |
| 235 | + return 0; |
| 236 | +} |
| 237 | + |
| 238 | +int |
| 239 | +PyList_Append(PyObject *op, PyObject *newitem) |
| 240 | +{ |
| 241 | + if (PyList_Check(op) && (newitem != NULL)) |
| 242 | + return app1((PyListObject *)op, newitem); |
| 243 | + PyErr_BadInternalCall(); |
| 244 | + return -1; |
| 245 | +} |
| 246 | +``` |
| 247 | + |
| 248 | +## 删除元素 |
| 249 | + |
| 250 | +```C |
| 251 | +static PyObject * |
| 252 | +listremove(PyListObject *self, PyObject *v) |
| 253 | +{ |
| 254 | + Py_ssize_t i; |
| 255 | + |
| 256 | + for (i = 0; i < self->ob_size; i++) { |
| 257 | + int cmp = PyObject_RichCompareBool(self->ob_item[i], v, Py_EQ); |
| 258 | + if (cmp> 0) { |
| 259 | + if (list_ass_slice(self, i, i+1, |
| 260 | + (PyObject *)NULL) == 0) |
| 261 | + Py_RETURN_NONE; |
| 262 | + return NULL; |
| 263 | + } |
| 264 | + else if (cmp < 0) |
| 265 | + return NULL; |
| 266 | + } |
| 267 | + PyErr_SetString(PyExc_ValueError, "list.remove(x): x not in list"); |
| 268 | + return NULL; |
| 269 | +} |
| 270 | +``` |
| 271 | + |
| 272 | +关键的是 list_ass_slice 函数,具体看源码,这里就粘贴了。 |
| 273 | + |
| 274 | +## PyListObject 对象缓冲池 free_lists |
| 275 | + |
| 276 | +当销毁 PyListObject 对象时,只释放了 PyListObject 关联的 PyObject* 数组占有的内存,PyListObject 本身的内存没有释放,而是缓存在 free_lists 中。当下次创建 PyListObject 对象时,可以直接利用 free_lists 中的 PyListObject。 |
| 277 | + |
| 278 | + |
| 279 | + |
| 280 | + |
| 281 | + |
0 commit comments