1 概述

最近研究了C++11的并发编程的线程/互斥/锁/条件变量，利用互斥/锁/条件变量实现一个支持多线程并发的环形队列，队列大小通过模板参数传递。
环形队列是一个模板类，有两个模块参数，参数1是元素类型，参数2是队列大小，默认是10。入队操作如果队列满阻塞，出队操作如果队列为空则阻塞。
其类图为:

2 实现

#ifndef RING_QUEUE_H
#define RING_QUEUE_H
#include <mutex>
#include <condition_variable>
template<typename T, std::size_t N = 10>
class ring_queue
{
public:
    typedef T           value_type;
    typedef std::size_t size_type;
    typedef std::size_t pos_type;
    typedef typename std::unique_lock<std::mutex> lock_type;
    ring_queue() { static_assert(N != 0); }
    ring_queue(ring_queue const&) = delete;
    ring_queue(ring_queue&& ) = delete;
    ring_queue& operator = (ring_queue const&) = delete;
    ring_queue& operator = (ring_queue &&) = delete;

    size_type spaces() const { return N; }
    bool empty() const
    {
        lock_type lock(mutex_);
        return read_pos_ == write_pos_;
    }

    size_type size() const
    {
        lock_type lock(mutex_);
        return N - space_size_;
    }

    void push(value_type const& value)
    {
        {
            lock_type lock(mutex_);
            while(!space_size_)
                write_cv_.wait(lock);

            queue_[write_pos_] = value;
            --space_size_;
            write_pos_ = next_pos(write_pos_);
        }
        read_cv_.notify_one();
    }

    void push(value_type && value)
    {
        {
            lock_type lock(mutex_);
            while(!space_size_)
                write_cv_.wait(lock);
            
            queue_[write_pos_] = std::move(value);
            --space_size_;
            write_pos_ = next_pos(write_pos_);
        }
        read_cv_.notify_one();
    }

    value_type pop()
    {
        value_type value;
        {
            lock_type lock(mutex_);
            while(N == space_size_)
                read_cv_.wait(lock);
            
            value = std::move(queue_[read_pos_]);
            ++space_size_;
            read_pos_ = next_pos(read_pos_);
        }
        write_cv_.notify_one();
        return value;
    }

private:
    pos_type next_pos(pos_type pos) { return (pos + 1) % N; }
private:
    value_type queue_[N];
    pos_type read_pos_ = 0;
    pos_type write_pos_ = 0;
    size_type space_size_ = N;
    std::mutex mutex_;
    std::condition_variable write_cv_;
    std::condition_variable read_cv_;
};
#endif

说明:

• 实现利用了一个固定大小数组/一个读位置/一个写位置/互斥/写条件变量/读条件变量/空间大小变量。
• 两个入队接口:
  • push(T const&) 左值入队
  • push(T &&) 左值入队
• 一个出队接口
  • pop()

3 测试

基于cpptest的测试代码如下：

struct Function4RingQueue
{
    ring_queue<std::string, 2> queue;
    std::mutex mutex;
    int counter = 0;
    void consume1(size_t n)
    {
        std::cerr << "\n";
        for(size_t i = 0; i < n; ++i)
        {
            std::cerr << "I get a " << queue.pop() << std::endl;
            counter++;
        }
    }
    void consume2(size_t id)
    {
        std::string fruit = queue.pop();
        {
            std::unique_lock<std::mutex> lock(mutex);
            std::cerr << "\nI get a " << fruit << " in thread(" << id << ")" << std::endl;
            counter++;
        }
    }
    void product1(std::vector<std::string> & fruits)
    {
        for(auto const& fruit: fruits)
            queue.push(fruit + std::string(" pie"));
    }
    void product2(std::vector<std::string> & fruits)
    {
        for(auto const& fruit: fruits)
            queue.push(fruit);
    }
};
void RingQueueSuite::one_to_one()
{
    Function4RingQueue function;
    std::vector<std::string> fruits{"Apple", "Banana", "Pear", "Plum", "Pineapple"};
    std::thread threads[2];

    threads[0] = std::thread(&Function4RingQueue::product1, std::ref(function), std::ref(fruits));
    threads[1] = std::thread(&Function4RingQueue::consume1, std::ref(function), fruits.size());

    for(auto &thread : threads)
        thread.join();
    TEST_ASSERT_EQUALS(fruits.size(), function.counter)

    function.counter = 0;
    threads[0] = std::thread(&Function4RingQueue::product2, std::ref(function), std::ref(fruits));
    threads[1] = std::thread(&Function4RingQueue::consume1, std::ref(function), fruits.size());

    for(auto &thread : threads)
        thread.join();
    TEST_ASSERT_EQUALS(fruits.size(), function.counter)
}

void RingQueueSuite::one_to_multi()
{
    Function4RingQueue function;
    std::vector<std::string> fruits{"Apple", "Banana", "Pear", "Plum", "Pineapple"};
    std::thread product;
    std::vector<std::thread> consumes(fruits.size());

    for(size_t i = 0; i < consumes.size(); ++i)
        consumes[i] = std::thread(&Function4RingQueue::consume2, std::ref(function), i);
    product = std::thread(&Function4RingQueue::product1, std::ref(function), std::ref(fruits));
    
    product.join();
    for(auto &thread : consumes)
        thread.join();
    TEST_ASSERT_EQUALS(fruits.size(), function.counter)

    function.counter = 0;
    for(size_t i = 0; i < consumes.size(); ++i)
        consumes[i] = std::thread(&Function4RingQueue::consume2, std::ref(function), i);
    product = std::thread(&Function4RingQueue::product2, std::ref(function), std::ref(fruits));
    product.join();
    for(auto &thread : consumes)
        thread.join();
    TEST_ASSERT_EQUALS(fruits.size(), function.counter)
}

• 函数one_to_one测试一个生成者对应一个消费者。
• 函数one_to_multi测试一个生产者对应多个消费者。

4 运行

RingQueueSuite: 0/2
I get a Apple pie
I get a Banana pie
I get a Pear pie
I get a Plum pie
I get a Pineapple pie

I get a Apple
I get a Banana
I get a Pear
I get a Plum
I get a Pineapple
RingQueueSuite: 1/2
I get a Apple pie in thread(1)

I get a Banana pie in thread(0)

I get a Pear pie in thread(2)

I get a Plum pie in thread(4)

I get a Pineapple pie in thread(3)

I get a Apple in thread(0)

I get a Banana in thread(1)

I get a Plum in thread(3)

I get a Pear in thread(2)

I get a Pineapple in thread(4)
RingQueueSuite: 2/2, 100% correct in 0.007452 seconds
Total: 2 tests, 100% correct in 0.007452 seconds

分析：

• 从结果看入队顺序和出队顺序是一致的。