Modern C++ concurrency - Returning values from Threads - std::future

Introduction

In this lesson we will talk about a way of returning values from threads, more precisely we will talk about std::future which is a mechanism that C++ offers in order to perform asynchronous tasks and query for the result in the future.
A future represents an asynchronous task, i.e. an operation that runs in parallel to the current thread and which the latter can wait (if it needs to) until the former is ready.
You can use a future all the time you need a thread to wait for a one-off event to happen. The thread can check the status of the asynchronous operation by periodically polling the future while still performing other tasks, or it can just wait for the future to become ready.

Future

To better understand what a future is, imagine a scenario in which your algorithm has to perform three tasks i.e. T1, T2 and T3.
T1 and T2 can be executed independently from each other while T3 in order to be executed needs both T1 and T2 to be completed.
The thread responsible for the execution of such algorithm can asynchronously launch a task for the calculation of T1 while keeping itself busy with the task T2. Whenever it finishes T2 it can simply wait for the asynchronous task T1 to be ready and then proceed with T3.

std::async()

The asynchronous task can be created using std::async function which returns a future and has the following signature:

template< class Function, class... Args >

std::future<std::result_of_t<std::decay_t<Function>(std::decay_t<Args>...)>>
    async( Function&amp;&amp; f, Args&amp;&amp;... args );

As you can see std::async returns a future of a certain type which depends on the type of Function that we supply to async.

Given a Function f of type RetType(Type1, Type2,...Typen) when async is called as in the following: async(f , arg1, arg2,...,argn); it will return a future of type std::future<RetType>. It is not surprising if the future holds the returning value of the Function f.

Given a future you can retrieve its payload using the get() member function which will block the calling thread until the future is in the ready state (meaning that the async task is completed and the result is safely stored in the future).

An Example

Let's start with the code for the example described above:

#include <thread>
#include <future>
#include <iostream>
#include <chrono>

double T1(){
    std::cout << "T1 : start" << std::endl;
    std::this_thread::sleep_for(std::chrono::seconds(5));
    std::cout << "T1 : end" << std::endl;
    return 5.2;   
}

int T2(){
    std::cout << "T2 : start" << std::endl;
    std::this_thread::sleep_for(std::chrono::seconds(5));
    std::cout << "T2 : end" << std::endl;
    return 444;  
}

void T3(double arg1, int arg2){
    std::cout << "T3 : start" << std::endl;
    std::this_thread::sleep_for(std::chrono::seconds(1));
    std::cout << "T3 : end" << std::endl;
}

int main(){
   
    auto start = std::chrono::high_resolution_clock::now();  

    std::cout << "I'm the main thread: start" << std::endl;
    {
    	auto future_t1 = std::async(T1);
   	 	const int res_t2 = T2();
    	const double res_t1 = future_t1.get();
    	T3(res_t1, res_t2);
    }
    std::cout << "I am the main thread: completed" << std::endl;
    
    auto end = std::chrono::high_resolution_clock::now();
    std::chrono::duration<double, std::milli> elapsed = end-start;
    std::cout << "Elapsed time " << elapsed.count() << " ms\n";
    return 0;
}

As you can see the main thread uses async to spawn an async task. std::async return immediately an instance of std::future<double> and the main continues executing the code for T2().
When T2() is complete, then the main simply waits for the async task to be completed using the std::future::get() function.
get() blocks the calling thread until the moment the future becomes ready. At that moment it returns the payload, which in this case is a double.
main() then proceeds with T3().

The following is the output that I get from the previous code compiled with clang++7.0.1 on my laptop:

I'm the main thread: start
T2 : start
T1 : start
T2 : end
T1 : end
T3 : start
T3 : end
I am the main thread: completed
Elapsed time 6000.65 ms

Please note since the total duration is ~6s this means that the std::async task actually runs in parallel with the main.

std::launch

The first thing that need to be noticed is that std::async(f), spawns an async task that might run in parallel! It is not guaranteed by the standard that f actually runs on a different thread, and it might be as well executed only when get() is called.
Thankfully the standard gives us also the possibility to control where the async tasks runs. std::async comes with an handy overload having the following signature:

template< class Function, class... Args >

std::future<std::result_of_t<std::decay_t<Function>(std::decay_t<Args>...)>>
    async( Function &amp; &amp; f, Args &amp; &amp;... args );

The first parameter of this overload is the enum std::launch which allows us to have control on the way the async tasks are executed. std::launch comes with two enum values:

std::launch::async: when used a new thread is launched to execute the task asynchronously
std::launch::deferred: when used the task is executed on the calling thread the first time its result is requested. This effectively mean that out task will be lazy evaluated (or using other words, executed in a call-by-need manner). The task start executing only when get() gets called on the corresponding future. This means that the task might not be executed at all.

Note that for the overloads of async that do not require a std::launch it is assumed that the launch policy is std::launch::async | std::launch::deferred (both option activated). In this case, the implementation has the right to choose which methods to use.

In fact when the example above is run using auto future_t1 = std::async(std::launch:deferred, T1); the output that I obtain is the following:

(21:57:13) ○ [knotman@archazzo] 
I'm the main thread: start
T2 : start
T2 : end
T1 : start
T1 : end
T3 : start
T3 : end
I am the main thread: completed
Elapsed time 11000.6 ms

Note that T1 does not start executing until get() is called and that the total duration is as expected 11s as if all functions ran serially.

std::future - other functions

get() is not the only useful function that std::future() offers.

The followings are the most important ones:

valid a boolean function returning true if the future refers to a valid task. It is always true unless you are working on a moved-on and/or default constructed future. Note that calling any of the following function other than:
- destructor
- move-assignment
- valid() itself

when valid()==false results in undefined behavior.

wait() similar to get() but it does not retrieve and consumes the payload of the future. It blocks the execution of the thread until the value is ready.
std:.future_status wait_for(const std::chrono::duration<Rep,Period>& timeout_duration ) which works similarly to wait() but it only waits for the time duration specified in its parameter. wait_for returns as soon as the value is ready or the timeout ends. It returns a future_status which is an enum specifying the status of the async task at that point. It can assume one of the following values:
- ready: the payload is ready
- deferred: as in launch, states that the value will be ready only when requested for it
- timeout which covers the case when wait_for waited for the whole duration of the timeout without ending up having the payload ready.

Conclusion

We have seen how to use the future object that allows for asynchronous computation in C++. During the next lesson we will discuss how we can use the information we discussed here in order to implement a quick-sort version that uses futures to speed up its execution.