[QUESTION] how to access raw data created in C++?

[IvanM76]

How is It possibile to create raw data (Dynamic allocated Array) in C++ and passing It as a variable to interpreter in order to modify it Python side but in a way that all the modifications are reflected also in c++ side? I mean something like boost::numpy::ndarray

[YannickJadoul]

There's py::array: https://pybind11.readthedocs.io/en/stable/advanced/pycpp/numpy.html

[IvanM76]

thankyou Yannick for the response. My problem is that using py::array the data are copied from c++ to the py::array object.
For example in this code:

py::scoped_interpreter guard{}; // start the interpreter and keep it alive
py::object main = py::module::import("__main__");
py::object globals = main.attr("__dict__");

double* myVect = new double[100];
memset(myVect, 0, sizeof(double) * 100);
myVect[0] = 1.0;
myVect[99] = 2.0;
size_t size = 100;
py::array pyVect((size_t)100, myVect);
globals["myArray"] = pyVect;
std::cout << myVect[0] << "    ---   " << myVect[99] << std::endl;
py::exec("print(myArray)");
py::exec("myArray = myArray[::-1]");
py::exec("print(myArray)");
std::cout << myVect[0] << "    ---   " << myVect[99] << std::endl;
delete[] myVect;

the array named myArray in python is reversed, but not myVect.
Is there a way to avoid the copy of the data and instead use directly the pointer created ?
thank you

EDIT (@YannickJadoul): Surround with ```

[YannickJadoul]

@IvanM76, have a look at #1042, for example. You need to provide a base parameter to the py::array constructor, such that pybind11 knows that you are taking care of the memory management and that something is keeping the data alive.

In your example, where your py::array outlives the call(s) into Python, you could just pass py::none or py::capsule(nullptr). This is mostly meant for the case where you return this array to a Python caller, and you want to be sure the NumPy array's data stays valid.

[bstaletic]

The question seems to have been answered. Closing.