Is it possible to have pure virtual slots?

SPlatten

I have a based class which creates several slots:

virtual void connected() = 0;
virtual void dataIn() = 0;
virtual void disconnected() = 0;

In my derived class I implement these slots:

void connected() override;
void dataIn() override;
void disconnected() override;

I'm asking if this is possible because since I've done this I'm getting some very odd behaviour which I cannot explain, execution is halting in internal Qt functions.

[Edit] I just read here:
https://stackoverflow.com/questions/2998216/does-qt-support-virtual-pure-slots

That the answer is yes, but the override keyword is not required and the driver functions should not appear under a public slots section...but having done that I'm still getting:

'function name' overrides a member function but is not marked 'override'

For each overridden slot, this does not explain the strange behaviour.

SPlatten

@SPlatten , I moved the socket connection from the constructor to its own function:

void clsModHelper::connectToXMLMPAM() {
    //Connect to the Application
    connectToHost(clsModHelper::mscpszHost, muint16Port);

    if ( waitForConnected(clsModHelper::mscintConnectionTimeout) != true ) {
        throw -1;
    }
}

main loops like this:

int main(int intArgc, char* parystrArgv[]) {
//Initialise the module                
    QApplication a(intArgc, parystrArgv);
    clsModFileIO obj(&a, intArgc, parystrArgv);

    try{
        obj.connectToXMLMPAM();
    } catch( exception& e) {
        std::cout << "Could not connect to XMLMPAM!";
        exit(EXIT_FAILURE);
    }
    return obj.intExitCode();
}

Now it works, thank you to everyone who helped me to resolve this issue.

SGaist

Hi,

What strange behavior do you get ?

SPlatten

@SGaist I had a break point on:

if ( waitForConnected(clsModHelper::mscintConnectionTimeout) ) {

I've set the mscintConnectionTimeout to 20000 ms. When I click on the step over icon, the debugger jumps into qobjectdefs_impl.h and stops on line 152:

(o->*f)((*reinterpret_cast<typename RemoveRef<SignalArgs>::Type *>(arg[II+1]))...), ApplyReturnValue<R>(arg[0]);

SGaist

What if you have a pure virtual with a default implementation ?

SPlatten

@SGaist, can you please explain what you mean? I'm not sure I understand.

Found this:
https://rvarago.medium.com/default-implementation-for-pure-virtual-functions-in-c-3c525cc820af

Will give it a try tomorrow.

SPlatten

@SGaist All the pure virtual functions have overloaded methods in the derived class or the derived class would be abstract. Presently these are very simple functions and just contain a qDebug() << "function name" entry.

SPlatten

@SGaist , updated....if I comment out the 3 connects which:

//    connect(this, &QTcpSocket::connected, this, &clsModHelper::connected);
//    connect(this, &QTcpSocket::disconnected, this, &clsModHelper::disconnected);
//    connect(this, &QIODevice::readyRead, this, &clsModHelper::dataIn);

It doesn't crash and the connection is made, waitForConnected doesn't crash and the application continues to where I expect.

To clarify, in the base class, the initial prototype for these slots:

    public slots:
        virtual void connected() = 0;
        virtual void dataIn() = 0;
        virtual void disconnected() = 0;

Then in my derived class:

    public slots:
        void connected() override;
        void dataIn() override;
        void disconnected() override;

The implementation, just to prove if works:

void clsModFileIO::connected() {
    //Start the module thread
    start();
}
void clsModFileIO::dataIn() {
    qdbg() << "dataIn";
}
void clsModFileIO::disconnected() {
    emit terminateModule();
}

If I then uncomment those connections and rebuild it does exactly the same its very repeatable.

Bonnie

I used to have pure virtual slots. It's quite a long time ago so I was using the old syntax to connect. Maybe you can try if that will work.
P.S. I think the "override" mark is irrelevant, you can add that. The SO post is so old that c++11 was not published at that time.

SPlatten

@Bonnie , thank you, the only reason I put in the override keyword is to get rid of the annoying yellow text appended at the end of each line suggesting it.

J.Hilk

@SPlatten is clsModHelper the base class of clsModFileIO?

SPlatten

@J-Hilk Yes

SPlatten

@Bonnie , just replaced connections with:

    connect(this, SIGNAL(connected), this, SLOT(connected));
    connect(this, SIGNAL(disconnected), this, SLOT(disconnected));
    connect(this, SIGNAL(readyRead), this, SLOT(dataIn));

It worked, no more crashing, thank you.

J.Hilk

@SPlatten ok, than this
connect(this, &QTcpSocket::connected, this, &clsModHelper::connected);

is explicitly invoking the pure virtual base function, thats bound to crash

connect(this, &QTcpSocket::connected, this, & clsModFileIO::connected);

should work fine

SPlatten

@J-Hilk , thank you, but isn't this the point of using a pure virtual slot, I want the actual derived slot to be worked out by the compiler so I can have any number of derived classes based the same pure virtual slot and then the actual slot gets called in its place?

SPlatten

@Bonnie , not so fast, I thought it worked because it didn't crash, but now I see in the Application Output:

2020-11-06 09:11:49.460550+0000 mdFileIO[4406:86031] QObject::connect: Parentheses expected, signal clsModHelper::connected in ../mdFileIO/clsModHelper.cpp:67
2020-11-06 09:11:49.460576+0000 mdFileIO[4406:86031] QObject::connect: Parentheses expected, signal clsModHelper::disconnected in ../mdFileIO/clsModHelper.cpp:68
2020-11-06 09:11:49.460583+0000 mdFileIO[4406:86031] QObject::connect: Parentheses expected, signal clsModHelper::readyRead in ../mdFileIO/clsModHelper.cpp:69

Bonnie

@SPlatten
The old syntax is not like that :)
As the output says you need parentheses like

    connect(this, SIGNAL(connected()), this, SLOT(connected()));
    connect(this, SIGNAL(disconnected()), this, SLOT(disconnected()));
    connect(this, SIGNAL(readyRead()), this, SLOT(dataIn()));

SPlatten

@Bonnie, its different, but still not right, now it stops in moc_clsModHelper.cpp on a line which reads:

case 4: _t->connected(); break;

The Application Output shows:

libc++abi.dylib: Pure virtual function called!

J.Hilk

@SPlatten sure, but you can always call the base implementation by implicitly invoking it and in many cases you want to do that.

take for example:

class BaseClass {
public:
    virtual void functionA() { qDebug () << "A"; }
};

class DerivedClass : public BaseClass{

public:
    void functionA() override {qDebug() << "B";}

};

class DerivedClass2: public BaseClass {
public:
    void functionA() override {
        BaseClass::functionA();
        qDebug() << "B";
    }
};

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

    DerivedClass c1;
    DerivedClass2 c2;

    c1.functionA();
    qDebug() << "----";
    c2.functionA();

}

this outputs:

B
----
A
B

where the 2nd class implicitly invokes the base implementation.

Bonnie

@SPlatten
Hmmm...can you post the full header file of clsModFileIO class?

SPlatten

@J-Hilk Its a pure virtual slot for the purpose of defining the structure and interface required to be implemented by the derived classes, thats why they have no implementation.