Hi, coroutine listers! The example is very awesome. but after reading though the source, i have a question: Is any limitation makes a push_coroutine must be caller and pull_coroutine must be the coro? This way, we have to yield out at the first line of coroutine function by calling caller(), just like the example showed: typedef coroutines::pull_coroutine<void> coro_pull; typedef coroutines::pull_coroutine<void> coro_push; auto asynchronous(F f) -> future<decltype(f())> { typedef promise<decltype(f())> CoroPromise; CoroPromise coro_promise; auto coro_future = coro_promise.get_future(); CurrentCoro current_coro = { make_shared<coro_pull>(std::bind([f](CoroPromise &coro_promise, coro_push &caller) { caller(); coro_stack.top().caller = &caller; coro_promise.set_value(f()); }, std::move(coro_promise), placeholders::_1); }; coro_stack.push(std::move(current_coro)); (*coro_stack.top().coro)(); coro_stack.pop(); return coro_future; } There is an extra call to caller() in coroutine function to jump out to prepare coro stack. Why not swap coro_pull and coro_push like this: auto asynchronous(F f) -> future<decltype(f())> { typedef promise<decltype(f())> CoroPromise; CoroPromise coro_promise; auto coro_future = coro_promise.get_future(); CurrentCoro current_coro = { make_shared<coro_push>(std::bind([f](CoroPromise &coro_promise, coro_pull &caller) { coro_stack.top().caller = &caller; coro_promise.set_value(f()); }, std::move(coro_promise), placeholders::_1); }; coro_stack.push(std::move(current_coro)); (*coro_stack.top().coro)(); coro_stack.pop(); return coro_future; } This still works on my machine, and has no extra "call to caller()", push_coroutine is not invoked until we explicitly run it (while pull_coroutine is invoked in constructor). So, my question is, s there any bad effects by using coroutines in this way? Thanks in advance!