Thread-safe C++ stack

Question

I am writing a multi-threaded app whereby different writers will be pushing objects onto a stack and readers pulling them off the stack (or at least pushing the pointer to an object)..

Are there any structures built-into C++ which can handle this without adding locking code etc.? If not, what about the Boost libraries?

EDIT:

I guess one reason I thought this could be built-in was that I was thinking purely in x86 space and thought that a PUSH/POP of pointers should be an atomic action at the instruction level.

I'm not sure if my initial hunch is true or not, but I guess this would not necessarily be true across all platforms. Though if running on x86, do you get atomic PUSHes and POPs to the stack and if so, does this essentially make it lock-free?

Answer 1

Yep: Boost.Thread is great, and should fit your needs very well. (These days, many people say that you could almost count Boost as built-in functionality.)

There is still no class that you could use out-of-the-box, but once you have the synchronization primitives at hand, it really is quite simple to implement your own thread-safe wrapper around, for example, std::stack. It could look something like this (not implementing every method...):

template <typename T> class MyThreadSafeStack {
  public:
    void push(const T& item) {
      boost::mutex::scoped_lock lock(m_mutex);
      m_stack.push(item);
    }
    void pop() {
      boost::mutex::scoped_lock lock(m_mutex);
      m_stack.pop();
    }
    T top() const { // note that we shouldn't return a reference,
                    // because another thread might pop() this
                    // object in the meanwhile
      boost::mutex::scoped_lock lock(m_mutex);
      return m_stack.top();
    }

  private:
    mutable boost::mutex m_mutex;
    std::stack<T> m_stack;
}    

---

If you are new to C++, please learn about RAII. Relevant to this case, Boost.Thread has the "scoped lock" classes to make it difficult to shoot yourself in the leg by forgetting to release a lock.

If you ever find yourself writing code like this:

void doStuff() {
  myLock.lock();
  if (!condition) {
    reportError();
    myLock.unlock();
    return;
  }
  try {
    doStuffThatMayThrow();
  }
  catch (std::exception& e) {
    myLock.unlock();
    throw e;
  }
  doMoreStuff();
  myLock.unlock();
}

, then you should just say no, and go RAII instead (syntax not directly from Boost):

void doStuff() {
  scoped_lock lock;
  if (!condition) {
    reportError();
    return;
  }
  doStuffThatMayThrow();
  doMoreStuff();
}

The point is that when the scoped_lock object goes out of scope, its destructor releases the resource -- in this case, the lock. This will always happen, no matter whether you exit the scope by throwing an exception, or by executing the odd return statement that your colleague sneakily added in the middle of your function, or simply by reaching the end of the function.