Re: Problem with memory management with builder.

[Kjell Ahlstedt <kjell ahlstedt bredband net>]

Hi Krzysztof,

I'm only 99% sure of this, but here's how I think it works:

When a top-level window (or any container) is deleted, its child widgets
are not unconditionally deleted. They are unreferenced, i.e. their
reference counts are decreased by 1, and only if a widget's ref count
reaches 0, the widget is deleted.

Now Gtk::Builder::get_widget(), as opposed to gtk_builder_get_object(),
increments the ref count of the fetched widget. (This is 100% sure.)

When you delete the window containing the button, the button is not
deleted, because "refBuilder->get_widget("button1", button);" has
increased the ref count of button1. When window1 is deleted, the
button's ref count decreases from 2 to 1, not from 1 to 0.

If you want to check my theory, you can read the ref count with
G_OBJECT(button->gobj())->ref_count, although you're not really supposed
to do that. ref_count is marked as private data in GObject, and I can
find no function you can use to read it.

Kjell

tor 2011-06-16 klockan 11:51 +0200 skrev Krzysztof Łochwinowicz:
>                 Hi Krzysztof,


>                 Deleting an already deleted object will not always be
>                 guaranteed to
>                 cause obvious problems like a segmentation fault.
>                 Sometimes, depending
>                 on what order objects are created in and whether you
>                 try to do anything
>                 after the delete, you may see no apparent problem. Try
>                 adding some heap
>                 activity between the delete window and the delete
>                 button; you may be
>                 able to provoke the crash that way. Or run it with
>                 valgrind, which
>                 should report the double-delete whether it would
>                 actually crash or not.
>                 
>                 Cheers,
>                 Rob
>                 
>         I do some heap activity but it does not change anything. I run
>         valgrind. See attached log. There is no invalid free.


>         -----------------------------------------------------------
>         Krzysztof Łochwinowicz
> 

> _______________________________________________
> gtkmm-list mailing list
> gtkmm-list gnome org
> http://mail.gnome.org/mailman/listinfo/gtkmm-list