Another asynchronous wait on a steady timer

This is a new version of an oldish Boost ASIO example of mine about asynchronously waiting on a timer, keeping advantage of C++11 features. If you are looking for something simpler, there's another post on the same matter but more focused on the bare ASIO functionality. Or you could go straight to the original source, the official tutorial on Boost.

Five years later, I found out that this post requires some adjustments. You could follow the link to its March 2018 version.

The main function of this example is spawning a new thread, that runs a function that does something indefinitely. But before creating the new thread, it would set an asynchronous timer, calling a function on its expiration that would cause the runner to terminate.

It makes sense to encapsulate both function in a single class, like this:
class MyJob
    MyJob(const MyJob&) = delete; // 1
    const MyJob& operator=(const MyJob& ) = delete;

    std::mutex mx_; // 2
    bool expired_;
    MyJob() : expired_(false) {}

    void log(const char* message) // 3
        std::unique_lock<std::mutex> lock(mx_);
        std::cout << message << std::endl;

    void timeout() // 4
        expired_ = true;

    void operator()() // 5
        for(int i = 0; !expired_; ++i)
            std::ostringstream os;
            os << '[' << i << ']';

1. I don't want an object of this class to be copyable, we'll see later why. So I remove from this class interface copy constructor and assignment operator, using the C++11 equals-delete marker.
2. There are two threads insisting on a shared resource (the standard output console), a mutex is needed to rule its access.
3. The shared resource is used in this function only. A lock on the member mutex takes care of protecting it.
4. When the timer expires, it is going to call this method.
5. This function contains the job that is going to run in another thread. Nothing fancy, actually. Just a forever loop with some logging and sleeping. The timeout is going to change the loop control variable, so that we can have a way out.

This is the user code for the class described above:
boost::asio::io_service io; // 1
boost::asio::steady_timer timer(io, std::chrono::seconds(3)); // 2

MyJob job; // 3
timer.async_wait([&job](const boost::system::error_code&) {
}); // 4

std::thread thread(std::ref(job)); // 5;
1. An ASIO I/O service is created.
2. I create a steady timer (that is just like an old deadline timer, but uses the C++11 chrono functionality) on the I/O service object.
3. An object that describes the job I want to run in another thread is instantiated.
4. When the timer expires, the passed lambda function is executed. It is an asynchronous call, so it returns immediately the control, that passed to the next instruction (5). The lambda would call the timeout() method on the job object, that has been captured by reference. Having defined the MyJob class as non-copyable, forgetting the ampersand, passing the job by value, results in a compiler error. Here I don't care about the error code parameter, that is set by ASIO to say if the timer has expired correctly or with an error. I just stop the job running. In a real-life usage a check would be expected.
5. Before running the I/O service, I create a thread on our job - again passed by reference, as the std::ref() shows. Again, trying to pass it by value would result in compiler errors.

Full C++ code for this example is on Github.

No comments:

Post a Comment