This library provides a set of easy to use coroutine primitives & utilities running on top of boost.asio. These will be of interest for applications that perform a lot of IO that want to not block unnecessarily, yet still want to have linear & readable code (i..e. avoid callbacks).
A minimum of Boost 1.82 is necessary as the ASIO in that version has needed support. C++ 20 is needed for C++ coroutines.
Below is a showcase of features, if you're new to coroutines or asynchronous programming, please see the primer.
The assumptions are:
io_contextis the execution_context of choice.- If
asio::io_contextis the executor, no more than one kernel thread executes within it at a time. - Eager execution is the way to go.
- A thread created with promise is only using promise stuff.
// a single threaded main running on an io_context
cobalt::main co_main(int argc, char ** argv)
{
// wrapper around asio::steady_timer
asio::steady_timer tim{co_await cobalt::this_coro::executor};
dt.expires_after(std::chrono::milliseconds(100));
co_await tim.async_wait(cobalt::use_op);
co_return 0;
}That is, main runs on a single threaded io_context.
It also hooks up signals, so that things like Ctrl+C get forwarded as cancellations automatically
Alternatively, run can be used manually.
cobalt::task<int> main_func()
{
asio::steady_timer tim{co_await cobalt::this_coro::executor};
dt.expires_after(std::chrono::milliseconds(100));
co_await tim.async_wait(cobalt::use_op);
co_return 0;
}
int main(int argc, char ** argv)
{
return run(main_func());
}The core primitive for creating your own functions is cobalt::promise<T>.
It is eager, i.e. it starts execution immediately, before you co_await.
cobalt::promise<void> test()
{
printf("test-1\n");
asio::steady_timer tim{co_await cobalt::this_coro::executor};
dt.expires_after(std::chrono::milliseconds(100));
co_await tim.async_wait(cobalt::use_op);
printf("test-2\n");
}
cobalt::main co_main(int argc, char ** argv)
{
printf("main-1\n");
auto tt = test();
printf("main-2\n");
co_await tt;
printf("main-3\n");
return 0;
}The output of the above will be:
main-1
test-1
main-2
test-2
main-3Unlike ops, returned by .wait, the promise can be disregarded; disregarding the promise does not cancel it, but rather detaches is. This makes it easy to
spin up multiple tasks to run in parallel. In order to avoid accidental detaching the promise type uses nodiscard unless one uses + to detach it:
cobalt::promise<void> my_task();
cobalt::main co_main()
{
// warns & cancels the task
my_task();
// ok
+my_task();
co_return 0;
}A task is a lazy alternative to a promise, that can be spawned onto or co_awaited on another executor.
An cobalt::task can also be used with spawn to turn it into an asio operation.
A generator is a coroutine that produces a series of values instead of one, but otherwise similar to promise.
cobalt::generator<int> test()
{
printf("test-1\n");
co_yield 1;
printf("test-2\n");
co_yield 2;
printf("test-3\n");
co_return 3;
}
cobalt::main co_main(int argc, char ** argv)
{
printf("main-1\n");
auto tt = test();
printf("main-2\n");
i = co_await tt; // 1
printf("main-3: %d\n", i);
i = co_await tt; // 2
printf("main-4: %d\n", i);
i = co_await tt; // 3
printf("main-5: %d\n", i);
co_return 0;
}main-1
test-1
main-2
main-3: 1
test-2
main-4: 2
test-3
main-5: 3
Channels are modeled on golang; they are different from boost.asio channels in that they don't go through the executor. Instead they directly context switch when possible.
cobalt::promise<void> test(cobalt::channel<int> & chan)
{
printf("Reader 1: %d\n", co_await chan.read());
printf("Reader 2: %d\n", co_await chan.read());
printf("Reader 3: %d\n", co_await chan.read());
}
cobalt::main co_main(int argc, char ** argv)
{
cobalt::channel<int> chan{0u /* buffer size */};
auto p = test(chan);
printf("Writer 1\n");
co_await chan.write(10);
printf("Writer 2\n");
co_await chan.write(11);
printf("Writer 3\n");
co_await chan.write(12);
printf("Writer 4\n");
co_await p;
co_return 0u;
}Writer-1
Reader-1: 10
Writer-2
Reader-1: 11
Writer-3
Reader-1: 12
Writer-4
To make writing asio operations that have an early completion easier, cobalt has an op-helper:
template<typename Timer>
struct wait_op : cobalt::op<system::error_code> // enable_op is to use ADL
{
Timer & tim;
wait_op(Timer & tim) : tim(tim) {}
// this gets used to determine if it needs to suspend for the op
void ready(cobalt::handler<system::error_code> h)
{
if (tim.expiry() < Timer::clock_type::now())
h(system::error_code(asio::error::operation_aborted));
}
// this gets used to initiate the op if ti needs to suspend
void initiate(cobalt::completion_handler<system::error_code> complete)
{
tim.async_wait(std::move(complete));
}
};
cobalt::main co_main(int argc, char ** argv)
{
cobalt::steady_timer tim{co_await cobalt::this_coro::executor}; // already expired
co_await wait_op(tim); // will not suspend, since its ready
}
race let's you await multiple awaitables at once.
cobalt::promise<void> delay(int ms)
{
asio::steady_timer tim{co_await cobalt::this_coro::executor};
dt.expires_after(std::chrono::milliseconds(ms));
co_await tim.async_wait(cobalt::use_op);
}
cobalt::main co_main(int argc, char ** argv)
{
auto res = co_await race(delay(100), delay(50));
asert(res == 1); // delay(50) completes earlier, delay(100) is not cancelled
co_return 0u;
}