Numpy C扩展：访问PyArrayObject中的PyObject

Question

I created a pandas dataframe with pd.DataFrame({"A": ["dog", "cat"]}) and sent to ac extension the resulting numpy array. In C, I do:

PyArrayObject * pArray = (PyArrayObject *) pArrayObj;
PyArray_Descr * pDesc = PyArray_DTYPE(pArray);
int kind = pDesc->kind;
npy_intp sz = PyArray_DIM(pArray, 0);
PyObject * p0 = (PyObject *) PyArray_GETPTR1(pArray, 0);

In the above 'kind' contains 'O' (for object array) and sz = 2 (as expected). However p0 does not seem to access the first object:

(gdb) p *p0
$10 = {ob_refcnt = 140737227161872, ob_type = 0x7ffff06e8180}

Notice corrupted refcnt and unresolved ob_type (was expecting PyUnicode_Type).

Any suggestions to correctly obtain a pointer to each element?

I'm using numpy 1.14 on Centos 7, g++ 6.2.1

Answer 1

Figured out how to do it:

npy_intp stride = PyArray_STRIDE(pArray, 0);
const char * dataptr = PyArray_BYTES(pArray);
PyObject * p1 = PyArray_GETITEM(pArray, dataptr);
PyObject * p2 = PyArray_GETITEM(pArray, dataptr + stride);

Answer 2

I found this question because I had the same problem. Turns out the initial code was almost correct, just some more pointer magic required in the last line:

PyArrayObject * pArray = (PyArrayObject *) pArrayObj;
PyObject * p0 = *(PyObject **) PyArray_GETPTR1(pArray, 0);