Boost::thread as data member

We want to write a C++ stopwatch, a way counting seconds giving the user a way to start and stopping it. We'll do that by a class having a boost::thread as private data member, and letting the user interact with it through a couple of public methods that implement the required functionality.
This is the sort of class declaration that I am thinking about:
class StopWatch : boost::noncopyable

{
private:
int count_; // 1
bool terminate_; // 2
boost::mutex mx_; // 3
boost::thread th_; // 4
void run(); // 5
public:
StopWatch() : count_(0), terminate_(false) {}
void start();
int stop(); // 6
};

1. The current stopwatch value.
2. A flag to manage the user request to the stopwatch to stop. It is going to be shared by two threads, so it requires a mutex.
3. Used to let each thread to work safely on the object status.
4. This boost::thread would care about the counter management.
5. The local thread would run on this function.
6. This function stops the clock, and returns its current value.

Some TDD

Before writing the actual code, I designed a few test cases to help me to going through the development. I used Google Test to do that. If you are interested in the matter, you could have a look to a few posts I wrote, starting from the one about how to set it up.

Here is just the first test I wrote, it should be easy for the reader to figure out the other ones:
TEST(TestStopWatch, NormalBehaviour)

{
StopWatch sw; // 1
sw.start(); // 2
int delay = 3;
boost::this_thread::sleep(boost::posix_time::seconds(delay)); // 3
EXPECT_EQ(delay, sw.stop()); // 4
}

1. We create an instance of the class.
2. The stopwatch starts.
3. We let this thread sleep for a while ...
4. and then check that stopping the watch we get the expected value.

We should care about robustness for our code, so I have written another few tests to check what happens if I stop a watch already stopped, or I try to start it when already started (nothing).

One useful aspect of writing test cases in advance, is that the developer should think of details that could be slipped from the design. For instance, in this case I found out that originally I had forgot to determine how the stopwatch should react if started, stopped, and then restarted again. I pondered a few seconds on the matter, and in the end I have arbitrarily chosen to let the counting going on without resetting.

Public member functions

I have written only a simple default ctor that, beside setting the termination flag to false, set the counter to zero. It could be useful to let the user to initialize the counter with a specific value and, if you want, you can easily improve in this sense the class, adding the relative test cases. Notice that the class extends boost::noncopyable, so that copy ctor and assignment operator are undefined and inaccessible.

The start function is short and cool. Well, we could discuss about its coolness degree, but for sure it is short:
void StopWatch::start()

{
boost::lock_guard<boost::mutex> lock(mx_); // 1
if(th_.joinable()) // 2
return;
terminate_ = false; // 3
th_ = boost::thread(&StopWatch::run, this); // 4
}

1. We are about to change the object status, we want to do it in a safe way, so we go through the mutex.
2. If the thread is joinable, it it already been started. We don't want to do anything more.
3. The status flag is already set by the ctor, but we need to reset it in case of restarting.
4. We are calling here the boost::thread move copy ctor, that rely on the new C++11 (AKA C++0x) RValue reference concept. Maybe this line would have been more readable if written like:
boost::thread temp(&Counter::run, this);

th_ = temp.move();

In this way we explicitely show that firstly we create a temporary boost::thread object that runs on the run() method of the current object, then we assign it to the member boost::thread object.

If we run step by step the code, we could appreciate how the member boost::thread object is intially created in a not-a-thread status, its thread_info are set to nullptr, and after [4] its value is swapped with the one in the temporary boost::thread object.

Given the start() function its stop() companion comes quite natural:
int StopWatch::stop()

{
{ // 1
boost::lock_guard<boost::mutex> lock(mx_);
if(!th_.joinable()) // 2
return -1;
else // 3
terminate_ = true;
}
th_.join(); // 4
return count_;
}

1. We want to limit the locking on the mutex on this part of the function code, so we open a specific scope.
2. If the member boost::thread is not joinable, then doesn't make any sense stopping it, we return a negative value to signal that something unexpected happened.
3. Otherwise we mark that the user asked he wants the clock to stop.
4. We let the running thread to be completed, then return the current value for the counter.

Private member function

Here is the code actually executed by the worker thread, on user request:
void StopWatch::run()

{
bool terminate; // 1
do {
boost::this_thread::sleep(boost::posix_time::seconds(1)); // 2
std::cout << ++count_ << ' ';
{ // 3
boost::lock_guard<boost::mutex> lock(mx_);
terminate = terminate_;
}
} while(!terminate);
}

1. To minimize the span of the critical region, we use a local variable to check the object status.
2. This toy stopwatch counts seconds. The thread sleeps for one second and then increase the counter.
3. This is the critical section where we read the current value for the object status, and we copy it to the local variable we are about to use to check when we have to stop counting.

There was already a similar post, where a class is designed to allow a delayed startup of a thread, but I think that this current example could be a bit more interesting.

No comments:

Post a Comment