Why the python object doesn't move?

[leslie-fang-intel]

I am trying to write some C++ function for Pytorch and wrap with pybind11.
Here is a class I defined and exposed through pybind11:

• Pybind11 Class

  py::class_<runtime::TaskModule, std::shared_ptr<runtime::TaskModule>>(m, "TaskModule")
      .def(py::init([](const py::object& module) {
        return std::make_shared<runtime::TaskModule>(module);
      }))
      .def("submit", [](runtime::TaskModule& self, py::args& args, py::kwargs& kwargs) {
        self.submit(std::move(args), std::move(kwargs));
        return;
      });

• Python test

x = torch.rand(64, 64, 3, 3)
task = TaskModule(module)
task.submit(x)

print(x) ?=> **why here X still exist?**

Since in the submit, I am passing args by reference and doing the std::move() for args.
Here inside submit function, args should be not implemented after std::move.
But why in the python side, the tensor X still exist?

[leslie-fang-intel]

Hi @wjakob, Could you kindly help to take a look of this question? I see some similar answer but not exactly what I am confused.