How can I achieve something similar to a semaphore using boost in c++? [duplicate]

This question already has answers here: 开发者_StackOverflow社区 C++0x has no semaphores? How to synchronize threads? (12 answers) Closed 7 years ago.

I noticed that boost does not seem to support semaphores. What's the easiest way to achieve a similar effect?

---

This is one way of implementing a very simple semaphore using Boost.Thread. It's an inter-thread semaphore, not an interprocess one. No warranties implied, etc. - I haven't even compiled the code. It illustrates how mutexes and condition variables interact, and assumes a reasonably recent version of Boost.

Notice how the mutex and condition variable are "paired" - threads must have a lock to the mutex to wait on the condition variable, and re-acquire the lock when they're woken up. Also, the code that changes the data needs to explicitly wake up other code that might be waiting. This means that the mutex, condition variable, data, and the condition(s) that cause the wakeup, are all closely coupled. The tight coupling also means that the data, mutex, and condition variable should be encapsulated if possible - any external modification can break the code in strange ways, including deadlocks, missed wakeups, and other strange bugs.

All this is really meant as a complement to Vlad Lazarenko's answer - understanding the theory and principles are at least as important as having "working" code, in multi-threaded programming.

#include <boost/thread/condition_variable.hpp>
#include <boost/thread/mutex.hpp>    
#include <boost/thread/lock_types.hpp>


class semaphore
{
    //The current semaphore count.
    unsigned int count_;

    //mutex_ protects count_.
    //Any code that reads or writes the count_ data must hold a lock on
    //the mutex.
    boost::mutex mutex_;

    //Code that increments count_ must notify the condition variable.
    boost::condition_variable condition_;

public:
    explicit semaphore(unsigned int initial_count) 
       : count_(initial_count),
         mutex_(), 
         condition_()
    {
    }

    unsigned int get_count() //for debugging/testing only
    {
        //The "lock" object locks the mutex when it's constructed,
        //and unlocks it when it's destroyed.
        boost::unique_lock<boost::mutex> lock(mutex_);
        return count_;
    }

    void signal() //called "release" in Java
    {
        boost::unique_lock<boost::mutex> lock(mutex_);

        ++count_;

        //Wake up any waiting threads. 
        //Always do this, even if count_ wasn't 0 on entry. 
        //Otherwise, we might not wake up enough waiting threads if we 
        //get a number of signal() calls in a row.
        condition_.notify_one(); 
    }

    void wait() //called "acquire" in Java
    {
        boost::unique_lock<boost::mutex> lock(mutex_);
        while (count_ == 0)
        {
             condition_.wait(lock);
        }
        --count_;
    }

};

---

You either need Boost Interprocess semaphore or Boost Thread synchronization primitives.

Mutex/Lock and condition are primitives that are commonly used to synchronize access to shared resources across multiple threads of a single process. There are exclusive, readers-writer and recursive/reentrant types of mutexes. Mutex, in other words, is an exclusive lock. Condition is used to achieve atomicity when you need to unlock the mutex and wait for object to change. When you start waiting on a condition, it unlocks the mutex and guarantees than unlock + call to wait is atomic and no other threads can modify a resource between those two operations.

Semaphore, on another case, is a mix of condition and mutex, and is used for exactly the same purpose but to synchronize access across processes.

See Mutex vs Semaphore.

There is also such thing as non-blocking/lock-free synchronization that is becoming very popular these days. I personally use it in high-frequency trading applications when amount of data is relatively very large and low latency does matter a lot.

In your case, I assume 5 philosophers can have a dinner inside a single process with 5 threads. In that case you have to use a mutex, not a semaphore. You might or might not use condition though. It depends on what exactly and how exactly you want to implement that dining procedure.

I am not sure how to describe it better as I will end up writing a book about it. So I'd recommend you find some book that is already written to understand the basic concepts. Once you know basics, you can use APIs/libraries/frameworks like POSIX threads, Boost Interprocess or Thread, ACE or even non-blocking algorithms to achieve what you want.

Good luck!

---

I made a semaphore class compatible with boosts TimedLockable concept, so it can be used with locks like boost::unique_lock<semaphore>. Its not a semaphore in a classical definition of one, but can be used as one. Still, hope it can be useful for someone.

Its somehow tested, but there is great possibility, that I did something wrong. Would be great if someone could prove it correctness.

class semaphore
{
private:
   semaphore(const semaphore & other);
   semaphore & operator = (const semaphore & other);

   boost::mutex _mutex;
   boost::condition_variable _condVar;
   size_t _count;

   class wait_predicate
   {
   private:
      const size_t & _countRef;
   public:
      wait_predicate(const size_t & countRef) : _countRef(countRef) {}
      bool operator()() { return _countRef > 0; }
   };

   // must be used inside a locked scope!
   inline wait_predicate getWaitPredicate() const 
   {
      return wait_predicate(_count);
   }

public:
   semaphore(size_t size): _count(size)
   {}

   void lock()
   {
      boost::unique_lock<boost::mutex> local_lock(_mutex);
      _condVar.wait(local_lock, getWaitPredicate());
      _count--;
   }

   void unlock()
   {
      boost::unique_lock<boost::mutex> local_lock(_mutex);
      _count++;
      _condVar.notify_one();
   }

   bool try_lock()
   {
      boost::unique_lock<boost::mutex> local_lock(_mutex);
      if (0 == _count)
         return false;

      _count--;
      return true;
   }

   template <typename Duration>
   bool try_lock_for(const Duration & duration)
   {
      boost::unique_lock<boost::mutex> local_lock(_mutex);
      if (!_condVar.wait_for(local_lock, duration, getWaitPredicate()))
         return false;

      _count--;
      return true;
   }

   template <class TimePoint>
   bool try_lock_until(const TimePoint & timePoint)
   {
      boost::unique_lock<boost::mutex> local_lock(_mutex);
      if (!_condVar.wait_until(local_lock, timePoint, getWaitPredicate()))
         return false;

      _count--;
      return true;
   }

   template <class WaitCriteria>
   bool timed_lock(const WaitCriteria & criteria)
   {
      boost::unique_lock<boost::mutex> local_lock(_mutex);
      if (!_condVar.timed_wait(local_lock, criteria, getWaitPredicate()))
         return false;

      _count--;
      return true;
   }
};