[Java-gnome-developer] Re: Use of destroy()

[Laurent M. <la...@ao...>]

>>>>> "'Mark" == 'Mark Howard' <mh...@ti...> writes:

  Mark> On Sat, Oct 30, 2004 at 02:23:34PM +0200, Laurent Martelli
  Mark> wrote:

  Mark> - Java objects are GCd when they go out of scope, unless
  Mark> events have been registered. Therefore, c objects can exist
  Mark> without corresponding java objects.
  >>  I believe this not a problem, and anyway there's probably
  >> nothing that can be done against it since some gtk objects may
  >> create ither gtk objects on their own. It's all right as long
  >> they destroy them when they are destroyed themselves.

  Mark> I agree in this patricular case. But what about when we
  Mark> create a window, hide it and then the java object goes out of
  Mark> scope. The c objects are still three and we have no way of
  Mark> accessing them. This is a memory leak.

Unless we unref in finalize().

  >> By the way, when we want to get a reference on such an object
  >> from java, we seem to always get an instance of
  >> org.gnu.gtk.Widget. It would be nice to get the actual instance
  >> so that more useful things can be done with the widget. I'll fill
  >> a bug report against that.

  Mark> I thought there already was a bug about this.

Is it http://bugzilla.gnome.org/show_bug.cgi?id=155075 ?

It's not only a glade problem. Bin.getChild() is also affected.

  Mark> Not sure. I agree, libglade.getWidget should do an appropriate
  Mark> cast. I'm not sure how simple that would be though.

  >> What about the finalize method which is called when a java object
  >> is garbage collected ? I think should call g_object_ref() when a
  >> java object is created so that we are pretty the C object won't
  >> be destroyed as long as the java object exists, and
  >> g_object_unref() in finalize(). This way, a C object would
  >> destroyed when no java objects references it and it's not
  >> referenced by a any C object either.

  Mark> When we add event handlers, we create references from the c
  Mark> object to the java object. I think we would almost always get
  Mark> the unfortunate case where the c object and java object
  Mark> reference each other, but nothing else references
  Mark> them. i.e. a memory leak

I had not thought of that. But the GC is supposed to also work in the
case of circular references. Unless the the reference create in C is a
root object for the GC.

But it would seem normal that the programmer must unregister from
events when a widget is not used anymore and is meant to be garbaged. 

  Mark> WeakReferences might seem the obvious answer, 

That's what I was about to say :-)

  Mark> but it isn't

:-(

  Mark> -- in this case we would always have to keep references to any
  Mark> java objects which we have defined events in. 

I must admit I don't see the point. To me, weak refs are really a
solution. I guess I do not see the whole picture. 

  Mark> What we really need is a weak reference which looks at both
  Mark> java references and g_object_refs. It might be possible for us
  Mark> to create this.

  >> This only problem would be if a C object is manually destroyed
  >> while there are still references on it (ref_count>0).
  Mark> This shouldn't happen.

If gtk does not do this, then we can assume it can't happen if we
remove destroy() from the java API.

-- 
Laurent Martelli
la...@ao...                                Java Aspect Components
http://www.aopsys.com/                          http://jac.objectweb.org