线程安全队列有问题吗？

Question

我正在尝试使用c ++中的pthreads编写一个线程安全的队列.我的程序有93%的时间都在工作.另外7%的时间它会吐出垃圾,或者似乎睡着了.我想知道我的队列中是否存在一些上下文切换会破坏它的缺陷？

// thread-safe queue
// inspired by http://msmvps.com/blogs/vandooren/archive/2007/01/05/creating-a-thread-safe-producer-consumer-queue-in-c-without-using-locks.aspx
// only works with one producer and one consumer
#include <pthread.h>
#include <exception>

template<class T>
class tsqueue
{
    private:
        volatile int m_ReadIndex, m_WriteIndex;
        volatile T *m_Data;
        volatile bool m_Done;
        const int m_Size;
        pthread_mutex_t m_ReadMutex, m_WriteMutex;
        pthread_cond_t m_ReadCond, m_WriteCond;
    public:
        tsqueue(const int &size);
        ~tsqueue();
        void push(const T &elem);
        T pop();
        void terminate();
        bool isDone() const;
};

template <class T>
tsqueue<T>::tsqueue(const int &size) : m_ReadIndex(0), m_WriteIndex(0), m_Size(size), m_Done(false) {
    m_Data = new T[size];
    pthread_mutex_init(&m_ReadMutex, NULL);
    pthread_mutex_init(&m_WriteMutex, NULL);
    pthread_cond_init(&m_WriteCond, NULL);
    pthread_cond_init(&m_WriteCond, NULL);
}

template <class T>
tsqueue<T>::~tsqueue() {
    delete[] m_Data;
    pthread_mutex_destroy(&m_ReadMutex);
    pthread_mutex_destroy(&m_WriteMutex);
    pthread_cond_destroy(&m_ReadCond);
    pthread_cond_destroy(&m_WriteCond);
}


template <class T>
void tsqueue<T>::push(const T &elem) {
    int next = (m_WriteIndex + 1) % m_Size;
    if(next == m_ReadIndex) {
        pthread_mutex_lock(&m_WriteMutex);
        pthread_cond_wait(&m_WriteCond, &m_WriteMutex);
        pthread_mutex_unlock(&m_WriteMutex);
    }
    m_Data[m_WriteIndex] = elem;
    m_WriteIndex = next;
    pthread_cond_signal(&m_ReadCond);
}

template <class T>
T tsqueue<T>::pop() {
    if(m_ReadIndex == m_WriteIndex) {
        pthread_mutex_lock(&m_ReadMutex);
        pthread_cond_wait(&m_ReadCond, &m_ReadMutex);
        pthread_mutex_unlock(&m_ReadMutex);
        if(m_Done && m_ReadIndex == m_WriteIndex) throw "queue empty and terminated";
    }
    int next = (m_ReadIndex +1) % m_Size;
    T elem = m_Data[m_ReadIndex];
    m_ReadIndex = next;
    pthread_cond_signal(&m_WriteCond);
    return elem;
}

template <class T>
void tsqueue<T>::terminate() {
    m_Done = true;
    pthread_cond_signal(&m_ReadCond);
}

template <class T>
bool tsqueue<T>::isDone() const {
    return (m_Done && m_ReadIndex == m_WriteIndex);
}

这可以这样使用:

// thread 1
while(cin.get(c)) {
    queue1.push(c);
}
queue1.terminate();


// thread 2
while(!queue1.isDone()) {
    try{ c = queue1.pop(); }
    catch(char const* str){break;}
    cout.put(c);
}

如果有人发现这个问题,请说出来:)

Answer 1

是的,这里肯定存在问题.您对队列成员变量的所有访问都发生在互斥锁之外.事实上,我并不完全确定你的互斥锁正在保护什么,因为它们只是在条件变量上等待.

此外,您的读者和编写器似乎总是以锁定步骤运行,从不允许队列超出一个元素的大小.