The key of this bug is:

Just after hide/destroy a window where pointer pressed (but not released), when pointer button release, the pointer position within/outside current process window area.

If pointer position within current process window area, then everything works fine.

If pointer position outside current process window area, then bug appears.

I think maybe this is a GTK bug, or Vision2 bug (have to call something to release pointer capture before hide/destroy window?).

After hide/destroy a window, if first call {EV_WIDGET}.enable_capture on next focused window (MAIN_WINDOW in the demo project), then {EV_WIDGET}.disable_capture on the window immediately. The bug disappears. So the hack of this bug is call following feature just after dialog hidden/destroyed.

enable_disable_capture_for_next_focused_widget
        -- Hack for bug#13108
    local
        l_env: EV_ENVIRONMENT
    do
        create l_env
        l_env.application.focus_in_actions.extend_kamikaze (agent (a_widget: EV_WIDGET)
                                                                    require
                                                                        a_widget_not_void: a_widget /= Void
                                                                    do
                                                                        if not a_widget.is_destroyed and then a_widget.is_displayed then
                                                                            a_widget.enable_capture
                                                                            a_widget.disable_capture
                                                                        end
                                                                    end)
    end