Re: Glib::wrap with copy=true semantics?

[Murray Cumming <murrayc murrayc com>]

On Fri, 2005-03-04 at 16:45 +0100, Murray Cumming wrote:
> On Thu, 2005-03-03 at 13:03 -0800, Philip Langdale wrote:
> > Hi,
> > 
> > I have a scenario here where there is a GtkAdjustment owned by
> > a C gtk widget and a Gtkmm gets the adjustment and then wraps it
> > in a Gtkmm wrapper. The wrapper should pass copy=true because it
> > isn't the owner of the underlying GtkObject. However, what I
> > observe is that the wrapper is created and it increments the
> > adjustment's refcount as expected but when the wrapper dies, it
> > does not decrement the refcount, and it calls gtk_object_destroy
> > which seems incorrect. Because the other ref holders aren't
> > expecting destroy to be called, they do not release their refs
> > and the object ends up being leaked when the other ref olders
> > do release their refs because there is still a +1 thanks to
> > the wrapper.
> > 
> > On the other hand, if I pass copy=false, it doesn't touch the
> > refcount at all and calls gtk_object_destroy when the wrapper
> > dies as you'd expect.
> > 
> > Is there a bug here or am I just expecting the wrong semantics?
> 
> If you tell us exactly what function you are talking about then we can
> investigate. A test case would be even better.

Also, in gtkmm, I don't think we ever do an extra refcount on
GtkAdjustments. We only do this in combination with Glib::RefPtr<>,
which we don't use with GtkObjects. Without Glib::RefPtr<> there is no
way to unref it again.

-- 
Murray Cumming
murrayc murrayc com
www.murrayc.com
www.openismus.com