Using gil_scoped_release and gil_scoped_acquire

[vakokako]

From documentation on gil locking, I couldn't get an understanding of how to use gil_scoped_release and gil_scoped_acquire correctly.

From this test I could understand that you have to gil_scoped_release once before running multiple threads and then gil_scoped_acquire in each thread and it works correctly in my application.

pybind11/tests/test_embed/test_interpreter.cpp

Lines 287 to 296 in 96b943b

	py::gil_scoped_release gil_release{};
	REQUIRE(has_pybind11_internals_static());
	auto threads = std::vector<std::thread>();
	for (auto i = 0; i < num_threads; ++i) {
	threads.emplace_back([&]() {
	py::gil_scoped_acquire gil{};
	locals["count"] = locals["count"].cast<int>() + 1;
	});
	}

My questions:

1. is there a guideline of how to use these two locks if I want to write c++ function f that executes some python code and make f be thread safe?
2. I already have one instance of scoped_interpreter that is created once for the whole application. Would it be a good idea to create gil_scoped_release once for the whole application scope and in any function that wants to run python code add gil_scoped_acquire?