Good way to extract numpy arrays in a tuple to std::vector

[faultdiagnosistoolbox]

I have just begun to experiment with pybind11 and have a couple of newbie questions.

I have a function in C++ that takes in a python function reference. The python function returns a tuple with two 1D numpy-arrays, e.g.

def f():
    return np.array([1, 2, 3]), np.array([4, 5, 6, 7])

My C++ function should run the python function and then extract the elements in the numpy arrays into two std::vector (this is required). This will be done fairly often, so I don't want to do unnecessary copies or steps.

Current assumptions: a) I know elements in the tuple are numpy-arrays, and b) I know the size of the elements (below type long).
This seems to work for me:

    auto z = py::reinterpret_borrow<py::tuple>(f());
    auto py_a = py::reinterpret_borrow<py::array_t<long>>(z[0]);
    auto py_b = py::reinterpret_borrow<py::array_t<long>>(z[1]);

    std::vector<long> a{};
    std::vector<long> b{};

    auto r = static_cast<long *>(py_a.request().ptr);
    std::copy(r, r + py_a.size(), std::back_inserter(a));
    r = static_cast<long *>(py_b.request().ptr);
    std::copy(r, r + py_b.size(), std::back_inserter(b));

Two questions:

1. Is this a reasonable approach, are there any problems, or how would you do it better?
2. Is it possible to get rid of assumption b?

Any comments or help is appreciated,

Erik