Possible memory leak in QNetworkAccessManager

[giorgio.cardarelli]

Hi, I'm running this simple code in valgrind and it finds some memory leaks. Where am I wrong? I'm using Qt5.0.2 shipped with Ubuntu Linux 13.04 64bit

As you can see, I do delete the reply using deleteLater and qApp is quit with a proper delay, just to be completely sure that everything went as expected.

Valgrind output is:
@
valgrind --leak-check=yes ./TestNetworkBug
...
==31550== LEAK SUMMARY:
==31550== definitely lost: 1,904 bytes in 26 blocks
==31550== indirectly lost: 2,601 bytes in 22 blocks
==31550== possibly lost: 0 bytes in 0 blocks
==31550== still reachable: 25,100 bytes in 605 blocks
==31550== suppressed: 0 bytes in 0 blocks
@

Thanks in advance,
Giorgio

@main.cpp
#include <QCoreApplication>
#include <QTimer>
#include <QObject>

#include "tester.h"

int main(int argc, char *argv[])
{
QCoreApplication a(argc, argv);

Tester tester(&a);

QObject::connect(&tester, SIGNAL(readyToQuit()), &a, SLOT(quit()));
QTimer::singleShot(1000, &tester, SLOT(get()));

return a.exec();

}@

@Tester.h
#ifndef TESTER_H
#define TESTER_H

#include <QObject>

class QNetworkAccessManager;
class QNetworkReply;

class Tester : public QObject
{
Q_OBJECT
public:
explicit Tester(QObject *parent = 0);
virtual ~Tester();

signals:
void readyToQuit();

public slots:
void get();
void finished(QNetworkReply *reply);

private:
QNetworkAccessManager * manager;
};

#endif // TESTER_H
@

@Tester.cpp
#include <QNetworkAccessManager>
#include <QNetworkReply>
#include <QtDebug>
#include <QTimer>

#include "tester.h"

Tester::Tester(QObject parent) :
QObject(parent)
{
manager = new QNetworkAccessManager(this);
connect(manager, SIGNAL(finished(QNetworkReply)), this, SLOT(finished(QNetworkReply*)));
}

Tester::~Tester()
{
qDebug() << "~Tester";
delete manager;
}

void Tester::get()
{
qDebug() << "Posting request";
manager->get( QNetworkRequest( QUrl(QString("http://www.google.it"))) );
}

void Tester::finished(QNetworkReply *reply)
{
qDebug() << "Getting reply";
reply->deleteLater();

QTimer::singleShot(1000, this, SIGNAL(readyToQuit()));

}
@

[sierdzio]

Valgrind is known to produce false positives for some Qt classes.

I'm not sure setting the application as the parent for your Tester object is a good idea.

Anyway, please try with newer Qt version.

[giorgio.cardarelli]

Hi,

same with Qt 5.2.1

@
==312== LEAK SUMMARY:
==312== definitely lost: 1,640 bytes in 24 blocks
==312== indirectly lost: 4,191 bytes in 41 blocks
==312== possibly lost: 0 bytes in 0 blocks
==312== still reachable: 26,700 bytes in 608 blocks
==312== suppressed: 0 bytes in 0 blocks
@

On a side note: why shouldn't be a good idea to set qApp as the parent for a object? In this particular case, it makes no difference to set or not to set a parent for my Tester instance. But, in this case:

@
Tester * tester = new Tester(&a);
@

the idea is to let qApp call tester's destructor.

Isn't it correct?

[sierdzio]

In your case, you are creating the Tester object on a stack. So I guessed it will be first deleted by QApplication (after quit() is called), and then the standard out-of-scope deletion will be attempted. So it might be that this is what Valgrind reports as a leak here.

Since you say that it makes no difference, I guess I was wrong.

[giorgio.cardarelli]

I think your guess is more than correct, but in real life it actually makes no difference.

On a side note, I would have expected a "double free or corruption" error, with a core-dump. But maybe I'm quite going off-topic here...