How to create numpy array from buffer_info without copy?

[hzhangxyz]

I want to create numpy array without copy data(does not own data), and I try the following code.

      return py::array_t<ScalarType>(py::buffer_info{
            s.data(),
            sizeof(ScalarType),
            py::format_descriptor<ScalarType>::format(),
            1,
            std::vector<Size>{s.size()},
            std::vector<Size>{sizeof(ScalarType)}});

But it copy data in fact.

Is there any way to create numpy array by array_t even in the situation of sharing data?

In my case, the strides could sometimes be very wired, so I have to use buffer_info, instead of setting dimensions and use c strides directly.

[hzhangxyz]

It seems adding the py::object owning the buffer as the second argument to the py::array_t constructor solve the question