[solved] MetaObject not referencing slots in derived class

[kshegunov]

@mrjj

I wonder if Derived class need

It does, so why do you wonder? :)

[mrjj]

@kshegunov

I often just wonder :)
nah, i never used
metaObject()->methodCount();

so was not 100% sure.

[kshegunov]

@mrjj
QObject::metaObject() returns the QMetaObject associated with the class (it's a static member, and the method provides virtualization over the class tree). If you don't have the Q_OBJECT macro, there's no staticMetaObject member for the class, and consequently there's no metaObject() override, thus you get the parent's staticMetaObject (or the last class that had the macro). Like in this case, there can't be reflection for the derived type, because the needed meta information isn't generated (i.e. the missing macro); the best you get is the information for the parent class.

Although, a need for this kind of introspection is rather rare ...

[mrjj]

@kshegunov
ahh, i never really examined the macro.
So thats how it works.
Its very cool.
I wish it was a pure c++ feature :)

[kshegunov]

@mrjj
Actually it is, somewhat. See here. The moc is needed to generate the meta information about the class, and the methods implementations, but for the static members or virtual functions declarations we have the good ol' preprocessor. ;)

[XavierLL]

Resolved!

I tried with the Q_OBJECT before but didn't rerun qmake... my fault :) and the errors I was obtaining made me think that maybe I could only use Q_OBJECT in classes directly derived from QObject, didn't think deeply about it.

I am just experimenting with all the metadata classes for a property based system UI.

Thanks a lot!

Xllr

[mrjj]

@kshegunov
oh. pretty neat.
I just wish it could list
members variables + type too :)

I never used the Qt metasystem. (directly)
Can moc be used to extract (plain) variables or does it has to be properties?

[XavierLL]

Not sure, because I have been working with this for 2 days, but I think they must be properties.

[kshegunov]

@mrjj said:

I just wish it could list
members variables + type too

Well, this wouldn't be very helpful, as you already know what member variables you have (and some of them may be private), or as in the usual case you only have a PIMPL pointer. You can however get the declared properties (such as declared with the Q_PROPERTY macro) and this is used extensively, e.g. in QML.

[mrjj]

@kshegunov
well it would be extremely helpful for creating boilerplate code for serialization and
trace systems & module tests and all kind of code gen.
Without adding ANYTHING to the source code as all serialization
frameworks i have seen does.

Maybe in c++32 :)

[kshegunov]

@mrjj said:

well it would be extremely helpful for creating boilerplate code for serialization

You have QDataStream for that. My suspicion is that moc's source will just explode if you start adding more and more parsing features. It already does a lot, e.g. RTTI without compiler RTTI and of course the signal-slot mechanism. If you need to wrap some boilerplate code you can always use a combination of the preprocessor with virtualization (similarly to what Qt does). I, personally, use a virtual stream operator for such things:

class MyClass
{
    friend QDataStream & operator << (QDataStream &, const MyClass &);
    friend QDataStream & operator >> (QDataStream &, MyClass &);

protected:
    virtual bool serialize(QDataStream &) = 0;
    virtual bool deserialize(QDataStream &) = 0;
}

inline QDataStream & operator << (QDataStream & out, const MyClass & obj)
{
    obj.serialize(out);
    return out;
}

inline QDataStream & operator >> (QDataStream & in, MyClass & obj)
{
    obj.deserialize(in);
    return in;
}

[mrjj]

@kshegunov
Well I guess it then again boils down to
out << var1 << var2 << var3 pr class which is what
i would like NOT to have to ever write. :)

So if could

for ( all member vars : curvar)
out << curvar;

make me very happy.

It seems Qt properties would allow such thing ? :)

[kshegunov]

@mrjj

It seems Qt properties would allow such thing ? :)

Yes, you can list them, but then again you have to declare them with Q_PROPERTY, and also you might want to save internal data that's not exposed through a property ... which could pose a significant problem.

I prefer the mentioned method, because I can delegate to the parent. Consider the following example:

class MyClassImpl : public MyClass
{
    // ...

protected:
    bool serialize(QDataStream & out) override
    {
        out << x;
        return true;
    }

    bool deserialize(QDataStream & in) override
    {
        in >> x;
        return true;
    }

private:
    int x;
}

class MyDerivedClassImpl : public MyClassImpl
{
    // ...

protected:
    bool serialize(QDataStream & out) override
    {
        MyClassImpl::serialize(out);
        out << y << str;
        return true;
    }

    bool deserialize(QDataStream & in) override
    {
        MyClassImpl::deserialize(in);
        out >> y >> str;
        return true;
    }

private:
    double y;
    QString str;
}

[mrjj]

@kshegunov
Yeah, 50% is not Qt enabled so would be an issue.
Im just daydreaming :)

Hmm, that is actually neat.