为什么我的 Python C 扩展会泄漏内存？
下面的函数接受一个 python 文件句柄，从文件中读取打包的二进制数据，创建一个 Python 字典并返回它。 如果我无休止地循环它，它将不断消耗内存。 我的引用计数出了什么问题？
static PyObject* __binParse_getDBHeader(PyObject *self, PyObject *args){
PyObject *o; //generic object
PyObject* pyDB = NULL; //this has to be a py file object
if (!PyArg_ParseTuple(args, "O", &pyDB)){
return NULL;
} else {
Py_INCREF(pyDB);
if (!PyFile_Check(pyDB)){
Py_DECREF(pyDB);
PyErr_SetString(PyExc_IOError, "argument 1 must be open file handle");
return NULL;
}
}
FILE *fhDB = PyFile_AsFile(pyDB);
long offset = 0;
DB_HEADER *pdbHeader = malloc(sizeof(DB_HEADER));
fseek(fhDB,offset,SEEK_SET); //at the beginning
fread(pdbHeader, 1, sizeof(DB_HEADER), fhDB );
if (ferror(fhDB)){
fclose(fhDB);
Py_DECREF(pyDB);
PyErr_SetString(PyExc_IOError, "failed reading database header");
return NULL;
}
Py_DECREF(pyDB);
PyObject *pyDBHeader = PyDict_New();
Py_INCREF(pyDBHeader);
o=PyInt_FromLong(pdbHeader->version_number);
PyDict_SetItemString(pyDBHeader, "version", o);
Py_DECREF(o);
PyObject *pyTimeList = PyList_New(0);
Py_INCREF(pyTimeList);
int i;
for (i=0; i<NUM_DRAWERS; i++){
//epochs
o=PyInt_FromLong(pdbHeader->last_good_test[i]);
PyList_Append(pyTimeList, o);
Py_DECREF(o);
}
PyDict_SetItemString(pyDBHeader, "lastTest", pyTimeList);
Py_DECREF(pyTimeList);
o=PyInt_FromLong(pdbHeader->temp);
PyDict_SetItemString(pyDBHeader, "temp", o);
Py_DECREF(o);
free(pdbHeader);
return (pyDBHeader);
}
The function below takes a python file handle, reads in packed binary data from the file, creates a Python dictionary and returns it. If I loop it endlessly, it'll continually consume RAM. What's wrong with my RefCounting?
static PyObject* __binParse_getDBHeader(PyObject *self, PyObject *args){
PyObject *o; //generic object
PyObject* pyDB = NULL; //this has to be a py file object
if (!PyArg_ParseTuple(args, "O", &pyDB)){
return NULL;
} else {
Py_INCREF(pyDB);
if (!PyFile_Check(pyDB)){
Py_DECREF(pyDB);
PyErr_SetString(PyExc_IOError, "argument 1 must be open file handle");
return NULL;
}
}
FILE *fhDB = PyFile_AsFile(pyDB);
long offset = 0;
DB_HEADER *pdbHeader = malloc(sizeof(DB_HEADER));
fseek(fhDB,offset,SEEK_SET); //at the beginning
fread(pdbHeader, 1, sizeof(DB_HEADER), fhDB );
if (ferror(fhDB)){
fclose(fhDB);
Py_DECREF(pyDB);
PyErr_SetString(PyExc_IOError, "failed reading database header");
return NULL;
}
Py_DECREF(pyDB);
PyObject *pyDBHeader = PyDict_New();
Py_INCREF(pyDBHeader);
o=PyInt_FromLong(pdbHeader->version_number);
PyDict_SetItemString(pyDBHeader, "version", o);
Py_DECREF(o);
PyObject *pyTimeList = PyList_New(0);
Py_INCREF(pyTimeList);
int i;
for (i=0; i<NUM_DRAWERS; i++){
//epochs
o=PyInt_FromLong(pdbHeader->last_good_test[i]);
PyList_Append(pyTimeList, o);
Py_DECREF(o);
}
PyDict_SetItemString(pyDBHeader, "lastTest", pyTimeList);
Py_DECREF(pyTimeList);
o=PyInt_FromLong(pdbHeader->temp);
PyDict_SetItemString(pyDBHeader, "temp", o);
Py_DECREF(o);
free(pdbHeader);
return (pyDBHeader);
}
如果你对这篇内容有疑问，欢迎到本站社区发帖提问 参与讨论，获取更多帮助，或者扫码二维码加入 Web 技术交流群。
绑定邮箱获取回复消息
由于您还没有绑定你的真实邮箱，如果其他用户或者作者回复了您的评论，将不能在第一时间通知您！
发布评论
评论（3）
PyDict_New()返回一个新引用，请检查文档 对于PyDict。 因此，如果您在创建后立即增加引用计数，则会有两个对它的引用。 当您将其作为结果值返回时，其中一个会传输给调用者，但另一个永远不会消失。您也不需要增加
pyTimeList。 当你创建它时它就是你的。 然而，你需要 decref 它，但你只 decref 它一次，所以它也被泄漏了。您也不需要在
pyDB上调用Py_INCREF。 它是借用的引用，只要您的函数不返回，它就不会消失，因为它仍然在较低的堆栈帧中被引用。仅当您想将引用保留在另一个结构中的某个位置时，您才需要增加引用计数。
比照。 API 文档
PyDict_New()returns a new reference, check the docs forPyDict. So if you increase the refcount immediately after creating it, you have two references to it. One is transferred to the caller when you return it as a result value, but the other one never goes aways.You also don't need to incref
pyTimeList. It's yours when you create it. However, you need to decref it, but you only decref it once, so it's leaked as well.You also don't need to call
Py_INCREFonpyDB. It's a borrowed reference and it won't go away as long as your function does not return, because it's still referenced in a lower stack frame.Only if you want to keep the reference in another structure somewhere, you need to increse the refcount.
Cf. the API docs
OT：连续调用
PyList_Append是一个性能问题。 由于您提前知道将获得多少结果，因此可以使用：注意，在调用
PyList_SET_ITEM后，您可能不会减少o的引用计数，因为它“窃取”一个参考。 检查文档。OT: Using successive calls to
PyList_Appendis a performance issue. Since you know how many results you'll get in advance, you can use:Observe that you may not decrease the refcount of
oafter callingPyList_SET_ITEM, because it "steals" a reference. Check the docs.我不知道Python-C。 然而，我对 COM 引用计数的经验表明，新创建的引用计数对象的引用计数为 1。 所以你的 Py_INCREF(pyDB) 在 PyArg_ParseTuple(args, "O", &pyDB) 和 PyObject *pyDBHeader = PyDict_New(); 之后 都是罪魁祸首。 他们的引用计数已经是 2。
I don't know about Python-C. However, My experience with COM reference counting says that a newly created reference-counted object has a reference count of 1. So your Py_INCREF(pyDB) after PyArg_ParseTuple(args, "O", &pyDB) and PyObject *pyDBHeader = PyDict_New(); are the culprit. Their reference counts are already 2.