2013-11-25 22:05, John Emmas skrev:
Hi
Kjell,
Many thanks for sending me your versions of those files. The good
news is that with only the exception of some commented sections,
your versions are identical to mine. Both versions exhibit the
same behaviour (and unfortunately, the same problems).
I'm wondering now if I've been fooling myself to think that this
SpinButton/Adjuster issue is somehow related to the previous one
that got solved. For various reasons it does look a bit similar -
but I'm now starting to think that the SpinButton/Adjuster problem
is actually a clearly defined issue in its own right. The reason
I'm wondering is that this sequence crashes:-
Create a SpinButton
Create an Adjustment
Attach the Adjustment to the SpinButton
Delete the Adjustment
Delete the SpinButton // Remember that the (deleted)
Adjustment is still attached !!
The previous sequence is the case I have tested
without a crash. You can see the result in the gdb output
gdblog1.txt that I sent to you.
When the destructor Gtk::Adjustment::~Adjustment() executes,
disconnect_cpp_wrapper() is called and it calls
g_object_steal_qdata((GObject*)gobj(), Glib::quark). Because
Glib::quark_ is removed,
Glib::ObjectBase::destroy_notify_callback_() and
Gtk::Object::destroy_notify_() is not called for the Adjustment.
In
https://mail.gnome.org/archives/gtkmm-list/2013-November/msg00052.html
you said that the crash occurs when gtk_spin_button_finalize()
calls gtk_spin_button_set_adjustment(GTK_SPIN_BUTTON(object),
NULL).
Let me guess: The crash occurs when
gtk_spin_button_set_adjustment() calls
g_object_unref(spin_button->adjustment). This is where the
GtkAdjustment looses its last reference, and it's finalized. Then
Glib::ObjectBase::destroy_notify_callback_()
will not be called, if Glib::quark_ has been removed. If
the GQuark functions don't work correctly for you, and
Glib::quark_ has not been removed, then Glib::ObjectBase::destroy_notify_callback_()
is called, and your program crashes.
Remember also that you reported strange contents
in a list of GQuarks in another test program in
https://mail.gnome.org/archives/gtkmm-list/2013-November/msg00022.html.
OTOH
this sequence doesn't crash:-
Create a SpinButton
Create an Adjustment
Attach the Adjustment to the SpinButton
Delete the SpinButton // Adjustment hasn't been deleted
yet !!
Delete the Adjustment
But wait a minute... during my tests I discovered some evidence
to suggest that this sequence wouldn't crash either:-
Create a SpinButton
Create an Adjustment
Attach the Adjustment to the SpinButton
Detach the Adjustment (using
'Gtk::SpinButton::unset_adjustment()')
Delete the Adjustment // These two deletions can now
Delete the SpinButton // get done (safely) in either
order
So I guess I'm questioning the safety of deleting a
Gtk::SpinButton whilst it has an (invalid) Gtk::Adjustment
attached. I don't know enough about the gtkmm spec to know
whether that should (or should not) be safe. It certainly isn't
safe at present, from what I can tell. Would it help for me to
try the same thing with a plain GTK+ app? If it's safe to do with
GTK+ I'd assume it should also be safe with gtkmm....
John
|
