There is indeed one last thing we should know about. Suppose, we write code as follows:

async(launch::async, f);
async(launch::async, g);

This might have the motivation of executing functions f and g (we do not care about their return values in this example) in concurrent threads and then doing different things at the same time. While running such code, we will notice that the code blocks on this calls, which is most probably not what we want.

So why does it block? Isn't async all about nonblocking asynchronous calls? Yes it is, but there is one special peculiarity: if a future was obtained from an async call with the launch::async policy, then its destructor performs a blocking wait.

This means that both the async calls from this short example are blocking because the lifetime of the futures they return ends in the same line! We can fix this by capturing their return values in variables with a longer lifetime.