QThread weird behaviour [closed]

问题:

I have a QThread running some code and I wanted it to do exit nicely and do some cleanings so the code goes as:

testdevice.h

class testDevice : public QThread 
{
    Q_OBJECT
... // some definitions 

protected:
    void run(void);

private:
    hid_device *handle;
    bool abort;

public:
    ~testDevice(void);
};

testdevice.cpp

testDevice::~testDevice(void)
{    
    mutex.lock();
    abort = true;
    mutex.unlock();
    wait();

    if (handle != NULL)
    {
        hid_close(handle);
    }

    hid_exit();
}

void testDevice::run(void)
{    
    while(true)
    {        
        mutex.lock();
        if(abort)
        {
           return;
        }
       mutex.unlock();
       if (_connected)
       {
        //... more code
       }
       else // Case device is not connected
       { 
       emit deviceConnected(_connected);// A breakpoint in here is not triggered after abort is true
       handle = hid_open(TEST_VID, TEST_PID, NULL); // this is where VS2010 shows the error "Unhandled exception at 0x71241a95 in project.exe: 0xC0000005: Access violation."
       //...code continues

The behaviour I was expecting was run() to be called and when ~testDevice() destructor is called abort is set to true and wait blocks the destructor, and runreturns and then the destructor continues.

What is happening is that run() is called and when I close the application the destructors ~testDevice() is called and abort is set to true and wait() returns and the destructor finishes...BUT then run() keeps running and I get Unhandled exception at 0x71241a95 in project.exe: 0xC0000005: Access violation.....I am running this as VS2010 debug, if I place breakpoints I get this all the time, but with no breakpoints I just got this once in a while...any clues?

Something weird is that when I place a breakpoint in abort = true; sometimes the first breakpoint that hits in there has a small blue '!' sign, and the second time it hits is a regular all red one. The same happens for the breakpoint at emit deviceConnected(_connected); but this is kinda random as well...I dont know what this '!' means...could all of this be a debug problem?

My general suspicion is that HIDAPI is running in a separate thread and it has its own bugs when someone call hid_exit(); maybe hid_read(); continues to run and since hid_quit(); was called, hid_read(); looses some pointers and doesnt close...just a wild shot and to confirm this I need some developer at the HIDAPI to come foward and say something...I hope they read this

回答1:

You seem to be using QThread in a custom way rather than the usual start() and quit() execution. The best would be that in an ideal world if you could refactor your code using quit() for quiting a QThread since that would be closer to the signal/slot mechanism of this QObject subclass.

However, I agree that it is not always possible. I cannot yet tell you based on your few lines you provided that it would make sense in this particular case, but at least I would encourage you to think about it.

Also, you could possibly use the QWaitCondition for waiting for something to happen rather than low-level or/and custom wait() calls.

As for quickly working around the issue you are facing if the issue at hand is compiler optimization of a boolean value like putting it into the cache, you could try having making it atomic by using the std::atomic (C++11) or QAtomicInt option. This will make sure that the compiler cannot optimize the reads and writes out.

You would write something like this with the std typedef:

std::atomic_bool abort;

If you need to support pre C++-11 compilers as well as Qt 4, you would use the Qt class like this:

QAtomicInt abort;