[Kde-bindings] playground/bindings/kimono

[Richard Dale]

On Sunday 25 February 2007, Richard Dale wrote:
> On Saturday 24 February 2007, Arno Rehn wrote:
> > Am Samstag, 24. Februar 2007 schrieb Richard Dale:
> > > Hi Arno
> > >
> > > On Saturday 24 February 2007, Richard Dale wrote:
> > > > On Saturday 24 February 2007, Arno Rehn wrote:
> > > > > * Added ReadProperty and WriteProperty classes. Now custom
> > > > > properties can be accessed via Property() and SetProperty() (and
> > > > > dbus?). However there seems to be a problem with marshalling, see
> > > > > FIXMEs in the code. The workaround should work as well as with
> > > > > marshalling, though.
> > > >
> > > > OK, I've just had a look at the code - I'm not sure you need to
> > > > involve QVariants. I'll try doing it with the normal marshalling and
> > > > see if I can get it to work.
> > > >
> > > > > * Autogeneration of Q_PROPERTY signatures and types now works
> > > > > properly. * Flags of properties should now be set correctly.
> > > >
> > > > Excellent! One more important feature to tick off on the TODO list.
> > >
> > > The good news is that properties can be set and got over dbus. The bad
> > > news is that I had to revert the recent changes to the way QMetaObjects
> > > are created to get it to work.
> >
> > Uhm, for me it works the way it is at the moment...
> >
> > > I'm not sure that I understand what bug you were
> > > fixing as I'd never noticed duplicate slot entries being created. In
> > > QObject.cs there is this condition:
> > >
> > >                 [SmokeMethod("metaObject", "()", "")]
> > >                 public virtual QMetaObject MetaObject() {
> > >                         if (Qyoto.IsSmokeClass(GetType())) {
> > >                                 return ProxyQObject().MetaObject();
> > >                         } else {
> > >                                 return Qyoto.GetMetaObject(this);
> > >                         }
> > >                 }
> > >
> > > Which should have catered for the problem of QMetaObjects being created
> > > twice for Smoke classes, rather than subclasses of Smoke classes, I
> > > would have thought.
> >
> > Ah yes, didn't see that. It would be OK if you access it via
> > MetaObject(). But consider the following: You have a class inheritance
> > hierachy like QWidget
> >    I- MainWidget
> >         I- SubWidget
> >
> > Now you call 'SubWidget.SuperClass().SuperClass()' - it should return the
> > MetaObject of QWidget. I checked that
> > with 'SubWidget.SuperClass().SuperClass().ClassName()', which should
> > therefore be 'QWidget', instead it was 'Qyoto.QWidget'. That confused me
> > a bit, so I
> > called 'SubWidget.SuperClass().SuperClass().SuperClass().ClassName()'.
> > That returned just 'QWidget'. With some extra debugging output I found
> > out that there is an extra MetaObject created for the Qyoto.QWidget class
> > which in turn inherits from the original C++ QWidget MetaObject. Hence
> > the duplicate signal/slot/property/whatever entries in the MetaObject.
>
> Yes, that sounds wrong, but I would be surprised if the slots/signal
> entries were duplicated. The code certainly didn't work correctly when it
> created QMetaObjects for classes that were subclassed more than once from a
> Smoke class like your heirachy above.
>
> > This problem was created by my own implementation of correct MetaObject
> > inheritance with more than one subclass, but the way it's done now is the
> > easiest way, I think.
>
> It is certainly tricky to get right - I'll have a look at your changes and
> think about it more.
>
> > As I said before, for me everything works perfectly,
> > could you discribe your problem any further?
>
> Yes, the complexpingpong example didn't work with the new QMetaObject code,
> although tutorial t14 did.
I've found out what the problem was. The new method depends on using
staticMetaObject, and the isn't one in the Smoke library for QObject. In
QtRuby I had to special case for that. So in make_metaObject() it tries to
find staticMetaObject for QObject, fails and returns 0, which causes a crash.
In complexping the Ping class is a direct subclass of QObject, and so that's
why it crashes.

        if (isSmokeClass) {
                Smoke::Index classId = qt_Smoke->idClass(className);
                Smoke::Index nameId =
qt_Smoke->idMethodName("staticMetaObject");
                Smoke::Index meth = qt_Smoke->findMethod(classId, nameId);
                if (meth <= 0) {
                        // Should never happen..
                        return 0;
                }

I'll do a patch of how I think this should work, but not check it in - just
post it on this list for discussion. The old way is wrong, and so is the new
one. We also need to make qt_metacall work when there is more than one
subclass of a Smoke Class.

-- Richard