getting that undefined vtable error again

mzimmers

Hi all -

This came up as the result of another thread, but I think it warrants its own thread. This code:

class MyNativeEventFilter: public QAbstractNativeEventFilter
{
    Q_OBJECT
private:
public:
     bool nativeEventFilter(const QByteArray &eventType, void *message, long *result)
    Q_DECL_OVERRIDE
    {
        Q_UNUSED(eventType) // unneeded as long as we're only running on Windows
        Q_UNUSED(result)
        MSG *msg = (MSG *)(message);

        if ((msg->message == WM_DEVICECHANGE)
           && (msg->wParam == DBT_DEVNODES_CHANGED))
        {
            emit notify();
        }
        return false;
    }
Q_SIGNALS:
    void notify();
};

gives me this

C:\wifibutton\build-wifibutton_utility-Desktop_Qt_5_15_0_MinGW_32_bit-Debug\..\utility\main.cpp:19: 
error: undefined reference to `vtable for MyNativeEventFilter'

The error goes away when I comment out the emit() statement. What causes this, and what am I doing wrong here?

Thanks...

EDIT:

Disregard the statement that it works without the emit(): I have to remove the signal declaration AND Q_OBJECT to get it to compile. So...do I correctly infer that QAbstractNativeEventFilter is not derived from QObject?

JonB

@mzimmers
It is not. Look at https://doc.qt.io/qt-5/qabstractnativeeventfilter.html.

kshegunov

Besides what @JonB sourced, and assuming the error persists (if we are talking *nix), put a non inlined trivial constructor before the first virtual and then recompile, should go away.

PS: Also for the love of god don't put virtual implementations inside the class declaration, the style being ugly and cluttering the interface notwithstanding, the compiler can't, and hence never will, inline them, so there's no performance gain even.

mzimmers

@kshegunov thanks for the suggestions. This isn't *nix, so I just created an intermediary class to handle the signal:

class MySignal: public QObject
{
    Q_OBJECT
public:
    void notify() { emit sendSignal(); }
Q_SIGNALS:
    void sendSignal();
};

class MyNativeEventFilter: public QAbstractNativeEventFilter
{
private:
    MySignal mySignal;
public:
    MyNativeEventFilter(QObject *parent);
    bool nativeEventFilter(const QByteArray &eventType, void *message, long *result)
    Q_DECL_OVERRIDE;
};
bool MyNativeEventFilter::nativeEventFilter(const QByteArray &eventType, void *message, long *result)
{
    Q_UNUSED(eventType) // unneeded as long as we're only running on Windows
    Q_UNUSED(result)
    MSG *msg = (MSG *)(message);

    if ((msg->message == WM_DEVICECHANGE)
            && (msg->wParam == DBT_DEVNODES_CHANGED))
    {
        mySignal.notify();
    }
    return false;
}

This appears to work, and I think is about as straightforward as I'm going to get it.

kshegunov

If you like, you could do this as well:

class MyNativeEventFilter : public QObject, public QAbstractNativeEventFilter
{
    Q_OBJECT
    // ...
};

mzimmers

@kshegunov that's a winner. Thanks!

It appears that I can derive additional information by making some other OS calls (eg RegisterDeviceNotificationA()), but honestly the MS documentation makes Faulkner look like a comic book, and I don't think it's worth it. When I get such an event, I'll just re-poll for devices.

JonB

@mzimmers said in getting that undefined vtable error again:

makes Faulkner look like a comic book

?

mzimmers

@JonB perhaps I should have said "Conrad." Baroque, complex, stentorian...while not completely unpleasant, virtually indecipherable.

JonB

@mzimmers
Oh! This is a bit advanced for me! I only recently discovered you are American. We don't do Faulkner. "Stentorian" --- do you really expect me to know what means without looking it up? :)