AMDG Lionel LE ROUX wrote:

To start with Boost, I wrote a (very simple) test_that_() function in C/Boost with a numeric::array& as argument. I compiled a .so file and I'm then able to seamlessly use this function from Python : that's for the successful part... Now I want to trace memory usage (and possible leaks) using Valgrind : then I don't use Python anymore and decide to stay in the C world : so I wrote a short test program in C which only includes a call to the function in the .so :

/int main(void) { Py_Initialize(); import_array();

npy_intp dim[]={1000}; PyArrayObject* X = (PyArrayObject*)PyArray_SimpleNew(1, dim, PyArray_FLOAT);

test_that_((numeric::array&)X);

Py_DECREF(X); Py_Finalize();

return 0; }/

I can compile and build the executable file with no error but when I run it : segmentation fault when I try to use PyArray_Size within the test_that() function, which only "link" a PyArrayObject to the numeric::array argument. The manipulation of the data seems to be OK but problem arises with PyArray_Size.

Well, looks like the double cast (PyArrayObject -> numeric::array -> PyArrayObject) is not valid. Could someone confirm (or not) my (very) bad coding ?

What is numeric::array? I'm not sure that you can safely cast a PyArrayObject to a numeric::array, but even if you could you can't cast a pointer to a PyArrayObject to a reference to a numeric::array. In Christ, Steven Watanabe