Question

While going through Qt code I had this basic question on the pimpl implementation.
As an example taking QWidget implementation.

QWidget        ---inherits---> QObject
   |                              |
 contains                      contains
   |                              |
  \ /                            \ /
QWidgetPrivate ---inherits---> QObjectPrivate

Now QWidget has two instances of QObjectPrivate (through inheritance and through containment).
Why was the implementation done in this way? Isn’t it an overhead to have two instances of the same object?

Answer 1

There aren’t two instances of QObjectPrivate when instantiating a QWidget. If you look closely at the header file for QObject, you’ll notice a protected constructor:

protected:
    QObject(QObjectPrivate &dd, QObject *parent = 0);

Which sets the instance of QObjectPrivate to that passed in via the protected constructor:

QObject::QObject(QObjectPrivate &dd, QObject *parent)
    : d_ptr(&dd)

This is the constructor called by all of the different QWidget constructors. The QWidget constructor passes in an instace of QWidgetPrivate which, as you noted, is a subclass of QObjectPrivate. Thus, only one instance of QObjectPrivate exists in QWidget.

Here’s the default QWidget constructor that illustrates this:

QWidget::QWidget(QWidget *parent, Qt::WindowFlags f)
    : QObject(*new QWidgetPrivate, 0), QPaintDevice()
{
    QT_TRY {
        d_func()->init(parent, f);
    } QT_CATCH(...) {
        QWidgetExceptionCleaner::cleanup(this, d_func());
        QT_RETHROW;
    }
}