I think, there is a problem with QNetworkReply and threads

[Wurgl]

However. I will drop this code totally and write my own class to access the server.

The other side (the server) is an Intel sa1110 Chip with 206 MHz (a Mobotix webcam) and my client runs on some Opteron 244 with 1.8GHZ. The Opteron needs 100% CPU-Time just for pure reading. This is far too slow code.

BTW: It works okay, if I create the three classes in main() and just do the retrieving in the subthread.

[Rembobo]

In first post example _reply object is read/written in h object thread (run method...) and read/written in main thread (dataAvailable slot) at same time (without synchronization. For example some pointer became invalid while still used as valid in other thread).

The Opteron needs 100% CPU-Time just for pure reading.

Try to wait until 100000 or more bytes arrive and then read them. The finished signal to get the last portion of bytes.
Try (in this case Hurtz slots will be executed in Hurtz thread)

Hurz::Hurz()
{
    moveToThread(this);
}

Hurz::Hurz()
{
    moveToThread(this);
}

To copy to clipboard, switch view to plain text mode 

Some example (it crashed before using moveToThread)

#include <QNetworkAccessManager>
#include <QNetworkReply>
#include <QCoreApplication>
#include <QThread>
#include <QEventLoop>
#include <QDebug>
#include <QTime>
 
class Hurz : public QThread
{
    Q_OBJECT
 
public slots:
    void dataAvailable();
    void slot_finished();
 
public:
    Hurz();
    void run();
 
private:
    QNetworkAccessManager *_manager;
    QNetworkReply *_reply;
    QNetworkRequest *_request;
    QTime m_time;
};
 
Hurz::Hurz()
{
    moveToThread(this);
    m_time = QTime::currentTime();
}
 
void Hurz::run()
{
    qDebug() << "Hurz::run thread id is " << QThread::currentThreadId();
    _manager = new QNetworkAccessManager();
 
    // you need to change the url to a different one which sends you "endless" data, eg. a file with almost gigabytes of data
    _request = new QNetworkRequest(QUrl("http://ftp.stack.nl/pub/users/dimitri/doxygen-1.6.2-setup.exe"));
    _reply = _manager->get(*_request);
    connect(_reply, SIGNAL(readyRead()), this, SLOT(dataAvailable()));
    connect(_reply, SIGNAL(finished()), this, SLOT(slot_finished()));
    exec();
}
 
void Hurz::dataAvailable()
{
    if (_reply->size() > 100000) {
        qDebug() << "Hurz::dataAvailable: " << "Thread id is" << QThread::currentThreadId();
        QByteArray arr = _reply->readAll();
        int ms = m_time.msecsTo(QTime::currentTime());
        m_time = QTime::currentTime();
        qDebug() << "    " << arr.size() << "in" << ms << "ms. " << (float)arr.size() / 1024 / ms * 1000 << "kB/s";
    }
}
 
void Hurz::slot_finished()
{
    qDebug() << "Hurz::slot_finished: " << "Thread id is" << QThread::currentThreadId();
    if (_reply->size() > 0) {
        QByteArray arr = _reply->readAll();
        int ms = m_time.msecsTo(QTime::currentTime());
        m_time = QTime::currentTime();
        qDebug() << "    " << arr.size() << "in" << ms << "ms. " << (float)arr.size() / 1024 / ms * 1000 << "kB/s";
    }
    quit();
}
 
int main(int argc, char **argv)
{
    QCoreApplication app(argc, argv);
 
    Hurz h;
    //QEventLoop evLoop;
    QObject::connect(&h, SIGNAL(finished()), &app, SLOT(quit()));
    h.start();
 
    //for(;;)
    //    evLoop.processEvents(QEventLoop::ExcludeUserInputEvents);
    //evLoop.exec(QEventLoop::ExcludeUserInputEvents);
    app.exec();
}
 
#include "main.moc"

#include <QNetworkAccessManager>
#include <QNetworkReply>
#include <QCoreApplication>
#include <QThread>
#include <QEventLoop>
#include <QDebug>
#include <QTime>

class Hurz : public QThread
{
    Q_OBJECT

public slots:
    void dataAvailable();
    void slot_finished();

public:
    Hurz();
    void run();

private:
    QNetworkAccessManager *_manager;
    QNetworkReply *_reply;
    QNetworkRequest *_request;
    QTime m_time;
};

Hurz::Hurz()
{
    moveToThread(this);
    m_time = QTime::currentTime();
}

void Hurz::run()
{
    qDebug() << "Hurz::run thread id is " << QThread::currentThreadId();
    _manager = new QNetworkAccessManager();

    // you need to change the url to a different one which sends you "endless" data, eg. a file with almost gigabytes of data
    _request = new QNetworkRequest(QUrl("http://ftp.stack.nl/pub/users/dimitri/doxygen-1.6.2-setup.exe"));
    _reply = _manager->get(*_request);
    connect(_reply, SIGNAL(readyRead()), this, SLOT(dataAvailable()));
    connect(_reply, SIGNAL(finished()), this, SLOT(slot_finished()));
    exec();
}

void Hurz::dataAvailable()
{
    if (_reply->size() > 100000) {
        qDebug() << "Hurz::dataAvailable: " << "Thread id is" << QThread::currentThreadId();
        QByteArray arr = _reply->readAll();
        int ms = m_time.msecsTo(QTime::currentTime());
        m_time = QTime::currentTime();
        qDebug() << "    " << arr.size() << "in" << ms << "ms. " << (float)arr.size() / 1024 / ms * 1000 << "kB/s";
    }
}

void Hurz::slot_finished()
{
    qDebug() << "Hurz::slot_finished: " << "Thread id is" << QThread::currentThreadId();
    if (_reply->size() > 0) {
        QByteArray arr = _reply->readAll();
        int ms = m_time.msecsTo(QTime::currentTime());
        m_time = QTime::currentTime();
        qDebug() << "    " << arr.size() << "in" << ms << "ms. " << (float)arr.size() / 1024 / ms * 1000 << "kB/s";
    }
    quit();
}

int main(int argc, char **argv)
{
    QCoreApplication app(argc, argv);

    Hurz h;
    //QEventLoop evLoop;
    QObject::connect(&h, SIGNAL(finished()), &app, SLOT(quit()));
    h.start();

    //for(;;)
    //    evLoop.processEvents(QEventLoop::ExcludeUserInputEvents);
    //evLoop.exec(QEventLoop::ExcludeUserInputEvents);
    app.exec();
}

#include "main.moc"

To copy to clipboard, switch view to plain text mode 

[Wurgl]

Thanks!

moveToThread() was the solution for both the crash and the time-eating. My selfwritten class uses about 2-3% of CPU, Qt uses about 7-8% which is okay.

However, it seems to be a little bit strange to me, I really expected, that a subclass of QThread already lives in that thread.

[wysota]

QThread object is the thread controller, not the thread itself. And if QThread lived in the thread it represents, what would happen to it once the thread was finished?

[Wurgl]

What do you need threads here for anyway?

Reason 1: The example ist just a dummy code. The real program has multiple stages with image processing, each of these stages may be time consuming. And since I get images in an endless stream, I avoid loosing any of these images. Each stage shall run in its own thread, stage 1 is grabbing the images, stage 2 is some analysing and filtering, stage 3 is some neuronal network processing, stage 4 is storing in the file system ... and a few more stages
Reason 2: Finally it will run on a machine with 8 cores, so why not using all of them?
Reason 3: Because it is cool, when *ix-utility top shows a cpu-usage greater than 100% :-)

QThread object is the thread controller, not the thread itself. And if QThread lived in the thread it represents, what would happen to it once the thread was finished?

So a clean implementation is to create some subclasses of QObjects in QThread::run() which do the real work and not using moveToThread()?

[faldzip]

However, it seems to be a little bit strange to me, I really expected, that a subclass of QThread already lives in that thread.

Because you haven't read the Qt Documentation about QThread class :P where it is clearly written than body of run() is finally in new created thread with QThread::start() method.