2

I'm trying to build a python extension which has a function returns a np.array. But it breaks when compiled with MinGW32 but works with VS2019. I tried to comment the code setting the flag NPY_ARRAY_OWNDATA, then MinGW32 and VS2019 both work, but lead a memory leak.

(I guess VS-built numpy cannot free memory which malloc by MinGW32.)

Is it possible using MinGW32 with numpy?

static PyObject *py_Decode(PyObject *self, PyObject *args)
{
  // Do something

  char *buffer = malloc(1024 * 1024 * 2);
  npy_intp dims[2] = {1024, 1024};
  PyObject *ret = PyArray_SimpleNewFromData(2, dims, NPY_SHORT, buffer);
  PyArray_ENABLEFLAGS((PyArrayObject *)ret, NPY_ARRAY_OWNDATA);  // When commented, mingw32 works, but memory leak.
  return ret; 
}

I'm using:

  • Python 3.8.3 32bit
  • numpy 1.18.5
  • VS2019 19.26.28806
  • mingw32-gcc 9.2.0
PX.Liu
  • 171
  • 1
  • 7
  • 1
    Related: https://stackoverflow.com/a/55959886/5769463 While the question is for Cython, the answer can be used for any C-extension: Use `PyArray_SetBaseObject`. – ead Aug 03 '20 at 06:01
  • @ead Thanks very much. It exactly sovled my problem. – PX.Liu Aug 04 '20 at 07:56

0 Answers0