std:call_once works like a barrier. It maintains access to a function (or a callable object). The first thread to reach it gets to execute the function. Until it has finished, any other thread that reaches the call_once line is blocked. After the first thread returns from the function, all other threads are released, too.

In order to organize this little choreography, a variable is needed from which the other threads can determine if they must wait and when they are released again. This is what our variable once_flag callflag; is for. Every call_once line also needs a once_flag instance as the argument prepending the function that shall be called only once.

Another nice detail is: If it happens, that the thread which is selected to execute the function in call_once fails because some exception is thrown, then the next thread is allowed to execute the function again. This happens in the hope that it will not throw an exception the next time.