[Development] Header diff for QtGui

[Frederik Gladhorn]

Since we were unsure about the accessibility classes we had QAccessible* 
marked as internal documentation wise, but sadly not moved the headers to be 
_p.h.
The plan is to make them public for 5.2.

Onsdag 26. juni 2013 14.44.48 skrev Thiago Macieira:
> One issue found:
> 
> On quarta-feira, 26 de junho de 2013 13.49.40, Thiago Macieira wrote:
> >  class Q_GUI_EXPORT QAccessibleInterface
> >  {
> > 
> > +protected:
> > +    virtual ~QAccessibleInterface();
> > +
> > 
> >  public:
> > -    virtual ~QAccessibleInterface() {}
> 
> SC/BC BREAK
> 
> Change in protection level is not acceptable (MSVC and Sun CC mangle the
> protection level). The constructors are public, so it's possible to derive
> from this class.
> 
> The destructor should go back to public.
> 
> This is also an SC break, because it makes it impossible to delete a
> QAccessibleInterface.

This was done intentionally since the class is currently internal and we want 
to prevent mis-use in the future (the interfaces are managed in a cache and 
should never be deleted by the user).
This breakage was done intentionally since so far no one should have used the 
class.
In my opinion making the dtor public is a step back and I seriously doubt that 
anyone is using this at the moment.

> 
> The rest are comments:
> >  class Q_GUI_EXPORT QAccessible
> > 
> > -#ifndef qdoc
> > -        :public QObject
> > -#endif
> > 
> >  {
> > 
> > -    Q_OBJECT
> > -    Q_ENUMS(Role Event State)
> > +    Q_GADGET
> > +    Q_ENUMS(Role Event)
> 
> Acceptable because QAccessible's only constructor is private. The change in
> class size and vtable will not affect anyone.
> 
> The change from Q_OBJECT to Q_GADGET means the virtual functions are gone
> now (qt_metacall, qt_metacast, metaObject), but no one could call them
> before either.
> 
> The tr() functions are gone too. That's the only question I have, because it
> was possible to call:
> 	QAccessible::tr("Hello");
That would be misusing the tr function since the context should be the actual 
context not a random one (almost as bad as QObject::tr which is a no-no unless 
the string is in QObject).

> 
> Opinions?
> 
> > --- a/src/gui/kernel/qwindow.h
> > +++ b/src/gui/kernel/qwindow.h
> > 
> > @@ -261,10 +286,12 @@ public Q_SLOTS:
> >      void setWidth(int arg);
> >      void setHeight(int arg);
> > 
> > -    void setMinimumWidth(int w);
> > -    void setMinimumHeight(int h);
> > -    void setMaximumWidth(int w);
> > -    void setMaximumHeight(int h);
> > +    Q_REVISION(1) void setMinimumWidth(int w);
> > +    Q_REVISION(1) void setMinimumHeight(int h);
> > +    Q_REVISION(1) void setMaximumWidth(int w);
> > +    Q_REVISION(1) void setMaximumHeight(int h);
> 
> The meta object system doesn't care about Q_REVISION. It will simply extract
> the information. So this is neither BIC nor SIC.
> 
> However, the information is relevant to QML. QML experts, please comment on
> whether the change above is safe.
If it was in the 5.0 release, I would say the properties should still be there 
and thus the Q_REVISION be removed since the classes otherwise suddenly lose 
the setters.

-- 
Best regards,
Frederik Gladhorn
Senior Software Engineer - Digia, Qt
Visit us on: http://qt.digia.com