shared_ptr 超出范围时 c++ 崩溃...在运行至少成功 100,000 次之后

Question

此代码在崩溃之前有运行 100,000 次（测试中运行宁 6 小时）。当崩溃发生时，它会在 shared_ptr 超出范围时发生......此函数的目的是填充传入的 shared_ptr 向量（在这种情况下），并按类型过滤消息。 ..所以并不是 historyQueue 中的所有消息都被添加。这个新填充的向量稍后用于发送这些 PM...这个函数也被其他线程调用，这就是为什么有一个 LockGuard，它是标准锁卫的类型定义（typedef std::lock_guard LockGuard）

bool MessageHistory::getMessages(vector< shared_ptr<ProtocolMessage> >& v,bool allMessages, bool playerFilter, int playerId, MessageFilter* filter)
{
    LockGuard lock(historyMutex);
    v.resize(historyQueue.size());
    unsigned count=0;

    for_vector(historyQueue,i)
    {   PM pm=historyQueue[i];//PM is a shared_ptr as well items in histortQueue
        const int uid=pm->GetPlayerDest();
        bool pmok =false;
        int pmtype=0;



        if(!pm->GetPid() || !pm->GetMid())
            continue;

        pmtype=(pm->GetPid() << 16) + pm->GetMid();



        if(filter && pmtype)
            pmok=filter->messageIsOk(pmtype);

        if ((allMessages || uid == -1 || (playerFilter && uid == playerId))
        && (filter == 0 || pmok))
        {   if(count>=v.size())
            {   break;
            }
            v[count++]=pm;
        }

    }//crash happens here after 100,000's of successful calling of this function
    v.resize(count);
    return true;
}

回溯：

Program terminated with signal 11, Segmentation fault.
#0  0x00007f74546b51b0 in ?? ()
(gdb) bt
#0  0x00007f74546b51b0 in ?? ()
#1  0x00000000005a7f41 in _M_release (this=0x7f7454204230)
    at /usr/include/c++/4.6/bits/shared_ptr_base.h:146
#2  ~__shared_count (this=0x7f747dff0d38, __in_chrg=<optimized out>)
    at /usr/include/c++/4.6/bits/shared_ptr_base.h:551
#3  ~__shared_ptr (this=0x7f747dff0d30, __in_chrg=<optimized out>)
    at /usr/include/c++/4.6/bits/shared_ptr_base.h:751
#4  ~shared_ptr (this=0x7f747dff0d30, __in_chrg=<optimized out>)
    at /usr/include/c++/4.6/bits/shared_ptr.h:93
#5  MessageHistory::getMessages (this=0x7e3cf18, v=..., allMessages=false, 
    playerFilter=true, playerId=-2141, filter=0x7e3cf88)
    at MessageHistory.cpp:177

调用getmessages的相关部分函数：

if (handIsActive.IsLocked() && history.size() > 0)
    {   vector< shared_ptr<ProtocolMessage> > lp;
        history.getMessages(lp,playerId,&noChatFilter);
        shared_ptr<ProtocolMessage> pm(new HandSoFar(lp));
        GameQueue::sendMessage(address, pm);
}   }

在此之后不再使用矢量 lp...

任何帮助将不胜感激....谢谢

Answer 1

我唯一的预感是你在其他地方有 UB。

它可能是您作为参考传入的矢量 (v) 上的竞赛。很容易忘记同步访问。

稍微跑题了：我可以建议稍微简化一下吗。

减少 code/separating 问题总是有助于保持复杂性，从而降低错误率：

bool getMessages(vector<PM> &v, Query const& query)
{
    LockGuard lock(historyMutex);

    v.clear();
    std::copy_if(historyQueue.begin(), historyQueue.end(), back_inserter(v), query);

    return true;
}

那里有很多 confusing/conflicting 东西：

条件

    if (count >= v.size()) {
        break;
    }

不可能是真的，因为前面 resize()

v.resize(count) 之后也是……这只是求 copy_if （实际上，更好的是 v.assign_if 但图书馆没有想到这一点。 Boost Range 会做到这一点）。

这里有太多的混合和冗余:

if (!pm->GetPid() || !pm->GetMid()) // 1.
    continue;

在这里，我们已经知道pmtype不能为零：

pmtype = (pm->GetPid() << 16) + pm->GetMid();

if (filter && pmtype) // pmtype cannot be zero 
    pmok = filter->messageIsOk(pmtype);

pmok 现在部分定义，仅当 filter!=0... 然后它在使用中变得毛茸茸：

if ((allMessages || uid == -1 || (playerFilter && uid == playerId)) && (filter == 0 || pmok)) {

显然更容易做到

bool pmok = !filter || filter->messageIsOk(pmtype);

if ((allMessages || uid == -1 || (playerFilter && uid == playerId)) && pmok) {

我认为 allMessages 没有按照它说的去做，这很可疑。它实际上做了一些与 playerFilter==false 现在实现的相反的事情。我认为你可能放错了括号，或你应该调用标志allUserIds？（我假设是后者）。
您可以使用 optional<int> playerId 简化界面。然而，显然 uid==-1 无论如何都是一个神奇的值，你为什么不把 -1 信号也设为 'no userid filter' 呢？
什么是for_vector（确定...写循环不是宏？再说一遍，copy_if就是干这个的）

使用这些调整，查询 class 可以像这样完全实现：

struct Query {
    bool allPlayers;
    int playerId;
    MessageFilter *filter;

    Query(bool allPlayers = true, int playerId = -1, MessageFilter* filter = nullptr) :
        allPlayers(allPlayers), playerId(playerId), filter(filter)
    {
    }

    bool operator()(PM const& pm) const {
        if (!(pm->GetPid() && pm->GetMid()))
            return false;

        int  pmtype = (pm->GetPid() << 16) + pm->GetMid();
        bool pmok   = !filter || filter->messageIsOk(pmtype);

        int  uid         = pm->GetPlayerDest();
        bool matchPlayer = uid == -1 || allPlayers || (playerId != -1 && uid == playerId);

        return pmok && matchPlayer;
    }
};

使用方式类似于

std::vector<MessageHistory::PM> v;
hist.getMessages(v, MessageHistory::Query { false, 42 }); // only uid 42; no MessageFilter

完整演示

Live On Coliru

#include <vector>
#include <mutex>
#include <memory>
#include <deque>
#include <algorithm>

using LockGuard = std::lock_guard<std::mutex>;
std::mutex historyMutex;

struct ProtocolMessage { //stub
    int GetPlayerDest() const { return 1; }
    int GetPid()        const { return 1; }
    int GetMid()        const { return 1; }
};

struct MessageFilter {
    virtual bool messageIsOk(int)  const { return true; }
    virtual ~MessageFilter() { }
};

struct MessageHistory 
{
    using PM = std::shared_ptr<ProtocolMessage>;

    struct Query {
        bool allPlayers;
        int playerId;
        MessageFilter *filter;

        Query(bool allPlayers = true, int playerId = -1, MessageFilter* filter = nullptr) :
            allPlayers(allPlayers), playerId(playerId), filter(filter)
        {
        }

        bool operator()(PM const& pm) const {
            if (!(pm->GetPid() && pm->GetMid()))
                return false;

            int  pmtype = (pm->GetPid() << 16) + pm->GetMid();
            bool pmok   = !filter || filter->messageIsOk(pmtype);

            int  uid         = pm->GetPlayerDest();
            bool matchPlayer = uid == -1 || allPlayers || (playerId != -1 && uid == playerId);

            return pmok && matchPlayer;
        }
    };

    bool getMessages(std::vector<PM> &v, Query const& query)
    {
        LockGuard lock(historyMutex);

        v.clear();
        std::copy_if(historyQueue.begin(), historyQueue.end(), back_inserter(v), query);

        return true;
    }

  private:
    std::deque<PM> historyQueue;
};

int main() {
    MessageHistory hist;

    std::vector<MessageHistory::PM> v;
    hist.getMessages(v, MessageHistory::Query { false, 42 }); // only uid 42; no MessageFilter
}

shared_ptr 超出范围时 c++ 崩溃...在运行至少成功 100,000 次之后

c++ Crash when shared_ptr goes out of scope...after running at least 100,000 times successfully

c++

gdb

shared-ptr

segmentation-fault

完整演示

shared_ptr 超出范围时 c++ 崩溃...在 运行 至少成功 100,000 次之后

c++ Crash when shared_ptr goes out of scope...after running at least 100,000 times successfully

c++

gdb

shared-ptr

segmentation-fault

完整演示

shared_ptr 超出范围时 c++ 崩溃...在运行至少成功 100,000 次之后