C++ std::async run on main thread

Question

IS there a way of running a function back on the main thread ?

So if I called a function via Async that downloaded a file and then parsed the data. It would then call a callback function which would run on my main UI thread and update the UI ?

I know threads are equal in the default C++ implementation so would I have to create a shared pointer to my main thread. How would I do this and pass the Async function not only the shared pointer to the main thread but also a pointer to the function I want to rrun on it and then run it on that main thread ?

Answer 1

I have been reading C++ Concurrency in Action and chapter four (AKA "The Chapter I Just Finished") describes a solution.

The Short Version

Have a shared std::deque<std::packaged_task<void()>> (or a similar sort of message/task queue). Your std::async -launched functions can push tasks to the queue, and your GUI thread can process them during its loop.

There Isn't Really a Long Version, but Here Is an Example

Shared Data

std::deque<std::packaged_task<void()>> tasks;
std::mutex tasks_mutex;
std::atomic<bool> gui_running;

The std::async Function

void one_off()
{
    std::packaged_task<void()> task(FUNCTION TO RUN ON GUI THREAD); //!!
    std::future<void> result = task.get_future();

    {
        std::lock_guard<std::mutex> lock(tasks_mutex);
        tasks.push_back(std::move(task));
    }

    // wait on result
    result.get();
}

The GUI Thread

void gui_thread()
{
    while (gui_running) {
        // process messages
        {
            std::unique_lock<std::mutex> lock(tasks_mutex);
            while (!tasks.empty()) {
                auto task(std::move(tasks.front()));
                tasks.pop_front();

                // unlock during the task
                lock.unlock();
                task();
                lock.lock();
            }
        }

        // "do gui work"
        std::this_thread::sleep_for(std::chrono::milliseconds(1000));
    }
}

Notes:

1. I am (always) learning, so there is a decent chance that my code is not great. The concept is at least sound though.

2. The destructor of the return value from std::async (a std::future<> ) will block until the operation launched with std::async completes (see std::async ), so waiting on the result of a task (as I do in my example) in one_off might not be a brilliant idea.

3. You may want to (I would, at least) create your own threadsafe MessageQueue type to improve code readability/maintainability/blah blah blah.

4. I swear there was one more thing I wanted to point out, but it escapes me right now.

Full Example

#include <atomic>
#include <chrono>
#include <deque>
#include <iostream>
#include <mutex>
#include <future>
#include <thread>


// shared stuff:
std::deque<std::packaged_task<void()>> tasks;
std::mutex tasks_mutex;
std::atomic<bool> gui_running;


void message()
{
   std::cout << std::this_thread::get_id() << std::endl;
}


void one_off()
{
    std::packaged_task<void()> task(message);
    std::future<void> result = task.get_future();

    {
        std::lock_guard<std::mutex> lock(tasks_mutex);
        tasks.push_back(std::move(task));
    }

    // wait on result
    result.get();
}


void gui_thread()
{
    std::cout << "gui thread: "; message();

    while (gui_running) {
        // process messages
        {
            std::unique_lock<std::mutex> lock(tasks_mutex);
            while (!tasks.empty()) {
                auto task(std::move(tasks.front()));
                tasks.pop_front();

                // unlock during the task
                lock.unlock();
                task();
                lock.lock();
            }
        }

        // "do gui work"
        std::this_thread::sleep_for(std::chrono::milliseconds(1000));
    }
}


int main()
{
    gui_running = true;

    std::cout << "main thread: "; message();
    std::thread gt(gui_thread);

    for (unsigned i = 0; i < 5; ++i) {
        // note:
        // these will be launched sequentially because result's
        // destructor will block until one_off completes
        auto result = std::async(std::launch::async, one_off);

        // maybe do something with result if it is not void
    }

    // the for loop will not complete until all the tasks have been
    // processed by gui_thread

    // ...

    // cleanup
    gui_running = false;
    gt.join();
}

Dat Output

$ ./messages
main thread: 140299226687296
gui thread: 140299210073856
140299210073856
140299210073856
140299210073856
140299210073856
140299210073856

Answer 2

Are you looking for std::launch::deferred ? Passing this parameter to std::async makes the task executed on the calling thread when the get() function is called for the first time.