[QUESTION] Binding an Eigen::ArrayxxD derived class

Hi,

I'm new to Pybind11, and I am trying to bind a custom C++ class derived from Eigen::Arrayxxd.
I understand that pybind provides automatic conversion of Eigen types to numpy arrays, but ideally, I'd like to keep the structure of my class while getting access to the stored data, be it a numpy array or something more custom made.

Here is a simplified version of the code :

class Grid : public Eigen::ArrayXXd
	{
	public:
		double Lx = 0., Ly = 0.;

		using Eigen::ArrayXXd::ArrayXXd;

		CORE_API static Grid set_Size(size_t x, size_t y, double lx, double ly, double val = 0.);
	};

and the binding :

void bind_grid(py::module& m)
{
py::class_<Grid> pyGrid(m, "Grid", "Grid");
	pyGrid.def(py::init<>(), "constructor")
		.def_static("set_Size", &Grid::set_Size)

}

If I use #include <pybind11/eigen.h> , I can create a Grid object in python, but as soon as I use a method that returns a Grid (for example set_Size()), it gets translated to a numpy array automatically.
If I remove the include, the object returned is a Grid, but I can't access the stored data anymore.

I could implement operators and accessors for the stored data when using solution 2, but I assume there is a more straight forward solution ?

Any help would be much appreciated !

Start with the shown Grid binding and compare its behavior with and without pybind11/eigen.h. Determine whether the requested preservation of Grid objects and access to their stored data is supported, and document or demonstrate the resulting binding approach.