<< Back to man.ChinaUnix.net

<< Back to man.ChinaUnix.net

Changes from 1.2 to 2.0

Changes from 1.2 to 2.0 — Incompatible changes made between version 1.2 and version 2.0

Incompatible changes from 1.2 to 2.0

The GNOME 2.0 porting guide on http://developer.gnome.org has some more detailed discussion of porting from 1.2 to 2.0. See the sections on GLib and GTK+.

GTK+ changed fairly substantially from version 1.2 to 2.0, much more so than from 1.0 to 1.2. Subsequent updates (possibilities are 2.0 to 2.2, 2.2 to 2.4, then to 3.0) will almost certainly be much, much smaller. Nonetheless, most programs written for 1.2 compile against 2.0 with few changes. The bulk of changes listed below are to obscure features or very specialized features, and compatibility interfaces exist whenever possible.

  • gtk_container_get_toplevels() was removed and replaced with gtk_window_list_toplevels(), which has different memory management on the return value (gtk_window_list_toplevels() copies the GList and also references each widget in the list, so you have to g_list_free() the list after first unref'ing each list member).

  • The gdk_time* functions have been removed. This functionality has been unused since the main loop was moved into GLib prior to 1.2.

  • The signature for GtkPrintFunc (used for gtk_item_factory_dump_items()) has been changed to take a const gchar * instead of gchar *, to match what we do for GLib, and other similar cases.

  • The detail arguments in the GtkStyleClass structure are now const gchar *.

  • gtk_paned_set_gutter_size() has been removed, since the small handle tab has been changed to include the entire area previously occupied by the gutter.

  • gtk_paned_set_handle_size() has been removed, in favor of a style property, since this is an option that only makes sense for themes to adjust.

  • GDK no longer selects OwnerGrabButtonMask for button presses. This means that the automatic grab that occurs when the user presses a button will have owner_events = FALSE, so all events are redirected to the grab window, even events that would normally go to other windows of the window's owner.

  • GtkColorSelectionDialog has now been moved into it's own set of files, gtkcolorseldialog.c and gtkcolorseldialog.h.

  • gtk_widget_shape_combine_mask() now keeps a reference count on the mask pixmap that is passed in.

  • The GtkPatternSpec has been moved to GLib as GPatternSpec, the pattern arguments to gtk_item_factory_dump_items() and gtk_item_factory_dump_rc() have thusly been changed to take a GPatternSpec instead of a GtkPatternSpec.

  • Type system changes:

    • GTK_TYPE_OBJECT is not a fundamental type anymore. Type checks of the style (GTK_FUNDAMENTAL_TYPE (some_type) == GTK_TYPE_OBJECT) will not work anymore. As a replacement, (GTK_TYPE_IS_OBJECT (some_type)) can be used now.

    • The following types vanished: GTK_TYPE_ARGS, GTK_TYPE_CALLBACK, GTK_TYPE_C_CALLBACK, GTK_TYPE_FOREIGN. The corresponding GtkArg fields and field access macros are also gone.


    • The type macros GTK_TYPE_MAKE() and GTK_TYPE_SEQNO() vanished, use of GTK_FUNDAMENTAL_TYPE() is discouraged. Instead, the corresponding GType API should be used: G_TYPE_FUNDAMENTAL(), G_TYPE_DERIVE_ID(), G_TYPE_BRANCH_SEQNO(). Note that the GLib type system doesn't build new type ids based on a global incremental sequential number anymore, but numbers new type ids sequentially per fundamental type branch.

    • The following type functions vanished/were replaced:

      Old FunctionReplacement
      gtk_type_query()being investigated

      Use of g_type_class_ref()/g_type_class_unref() and g_type_class_peek() is recommended over usage of gtk_type_class(). Use of g_type_register_static()/g_type_register_dynamic() is recommended over usage of gtk_type_unique().

  • Object system changes: GtkObject derives from GObject, so is not the basic object type anymore. This imposes the following source incompatible changes:

    • GtkObject has no klass field anymore, an object's class can be retrieved with the object's coresponding GTK_<OBJECT>_GET_CLASS (object) macro.

    • GtkObjectClass has no type field anymore, a class's type can be retrived with the GTK_CLASS_TYPE (class) macro.

    • GtkObjectClass does not introduce the finalize() and shutdown() methods anymore. While shutdown() is intended for GTK+ internal use only, finalize() is required by a variety of object implementations. GObjectClass.finalize should be overriden here, e.g.:

          static void gtk_label_finalize (GObject *gobject)
            GtkLabel *label = GTK_LABEL (gobject);
            G_OBJECT_CLASS (parent_class)->finalize (object);
          static void gtk_label_class_init (GtkLabelClass *class)
            GObjectClass *gobject_class = G_OBJECT_CLASS (class);
            gobject_class->finalize = gtk_label_finalize;

  • The GtkObject::destroy signal can now be emitted multiple times on an object. ::destroy implementations should check that make sure that they take this into account, by checking to make sure that resources are there before freeing them. For example:

        if (object->foo_data)
            g_free (object->foo_data);
            object->foo_data = NULL;

    Also, ::destroy implementations have to release object references that the object holds. Code in finalize implementations such as:

        if (object->adjustment)
            gtk_object_unref (object->adjustment);
            object->adjustment = NULL;

    have to be moved into the ::destroy implementations. The reason for doing this is that all object reference cycles should be broken at destruction time. Because the ::destroy signal can be emitted multiple times, it no longer makes sense to check if a widget has been destroyed using the GTK_OBJECT_DESTROYED() macro, and this macro has been removed. If catching destruction is still needed, it can be done with a signal connection to ::destroy.

  • Signal system changes: The GTK+ 2.0 signal system merely proxies the GSignal system now. For future usage, direct use of the GSignal API is recommended, this avoids significant performance hits where GtkArg structures have to be converted into GValues. For language bindings, GSignal+GClosure provide a much more flexible and convenient mechanism to hook into signal emissions or install class default handlers, so the old GtkSignal API for language bindings is not supported anymore.

    Functions that got removed in the GTK+ signal API: gtk_signal_n_emissions(), gtk_signal_n_emissions_by_name(), gtk_signal_set_funcs(), gtk_signal_handler_pending_by_id(), gtk_signal_add_emission_hook(), gtk_signal_add_emission_hook_full(), gtk_signal_remove_emission_hook(), gtk_signal_query(). Also, the GtkCallbackMarshal argument to gtk_signal_connect_full() is not supported anymore. For many of the removed functions, similar variants are available in the g_signal_* namespace. The GSignal system performs emissions in a slightly different manner than the old GtkSignal code. Signal handlers that are connected to signal "foo" on object "bar" while "foo" is being emitted, will not be called anymore during the emission they were connected within.

  • Inserting and deleting text in GtkEntry though functions such as gtk_entry_insert_text() now leave the cursor at its original position in the text instead of moving it to the location of the insertion/deletion.

  • The label field of GtkFrame widgets has been removed (as part of a change to allow arbitrary widgets in the title position). The text can now be retrieved with the new function gtk_frame_get_text().

  • The 'font' and 'font_set' declarations in RC files are now ignored. There is a new 'font_name' field that holds the string form of a Pango font.

  • A number of types in GDK have become subclasses of GObject. For the most part, this should not break anyone's code. However, it's now possible/encouraged to use g_object_ref()/g_object_unref() and other GObject features with these GDK types. The converted types are: GdkWindow, GdkDrawable, GdkPixmap, GdkImage, GdkGC, GdkDragContext, GdkColormap.

  • All drawables including pixmaps used to have a type tag, the GdkWindowType enumeration, which included GDK_WINDOW_PIXMAP. GdkWindowType is now a property of GdkWindow only, and there is no GDK_WINDOW_PIXMAP. You can use the GDK_IS_PIXMAP() macro to see if you have a pixmap, if you need to know that.

  • GtkStyle and GtkRcStyle are now subclasses of GObject as well. This requires fairly extensive changes to theme engines, but shouldn't affect most other code.

  • xthickness and ythickness have moved from GtkStyleClass to GtkStyle (from class to instance). This gives themes a bit more flexibility and is generally more of the Right Thing. You can trivially fix your code with s/style->klass->xthickness/style->xthickness/g and same for ythickness.

  • Some GtkStyle draw_* methods have been removed (cross, oval, ramp) and others have been added (expander, layout). This will require changes to theme engines.

  • If you were using private GDK types, they have been rearranged significantly. You shouldn't use private types. ;-)

  • The visual for a widget, and also the default visual is now derived from the colormap for the widget and the default colormap. gtk_widget_set_visual(), gtk_widget_set_default_visual(), gtk_widget_push_visual() and gtk_widget_pop_visual() now do nothing. Since the visual always had to match that of the colormap, it is safe to simply delete all references to these functions.

  • A number of functions in GDK have been renamed for consistency and clarity. #defines to provide backwards compatibility have been included, but can be disabled by defining GDK_DISABLE_DEPRECATED.

    Old functionDefined As

    (Note that g_object_ref() and g_object_unref() may be used for all of the above ref and unref functions.) gtk_widget_popup() was removed, it was only usable for GtkWindows, and there the same effect can be achieved by gtk_window_move() and gtk_widget_show().

  • gdk_pixmap_foreign_new() no longer calls XFreePixmap() on the pixmap when the GdkPixmap is finalized. This change corresponds to the behavior of gdk_window_foreign_new(), and fixes a lot of problems with code where the pixmap wasn't supposed to be freed. If XFreePixmap() is needed, it can be done using the destroy-notification facilities of g_object_set_data().

  • GtkProgress/GtkProgressBar had serious problems in GTK+ 1.2.

    • Only 3 or 4 functions are really needed for 95% of progress interfaces; GtkProgress/GtkProgressBar had about 25 functions, and didn't even include these 3 or 4.

    • In activity mode, the API involves setting the adjustment to any random value, just to have the side effect of calling the progress bar update function - the adjustment is totally ignored in activity mode.

    • You set the activity step as a pixel value, which means to set the activity step you basically need to connect to size_allocate.

    • There are ctree_set_expander_style()-functions, to randomly change look-and-feel for no good reason.

    • The split between GtkProgress and GtkProgressBar makes no sense to me whatsoever.

    This was a big wart on GTK+ and made people waste lots of time, both learning and using the interface. So, we have added what we feel is the correct API, and marked all the rest deprecated. However, the changes are 100% backward-compatible and should break no existing code. The following 5 functions are the new programming interface and you should consider changing your code to use them:

      void       gtk_progress_bar_pulse                (GtkProgressBar *pbar);
      void       gtk_progress_bar_set_text             (GtkProgressBar *pbar,
                                                        const gchar    *text);
      void       gtk_progress_bar_set_fraction         (GtkProgressBar *pbar,
                                                        gfloat          fraction);
      void       gtk_progress_bar_set_pulse_step       (GtkProgressBar *pbar,
                                                        gfloat          fraction);
      void       gtk_progress_bar_set_orientation      (GtkProgressBar *pbar,
    						    GtkProgressBarOrientation orientation);

  • The GtkNotebookPage structure has been removed from the public header files; this was never meant to be a public structure, and all functionality that could be done by accessing the struct fields of this structure should be accessible otherwise.

  • Negative values of the position parameter to gtk_notebook_reorder_child() now cause the page to be appended, not inserted at the beginning. (This gives consistency with gtk_box_reorder_child(), gtk_menu_reorder_child().)

  • GtkMenuPositionFunc has a new parameter push_in which controls how menus placed outside the screen is handled. If this is set to TRUE and part of the menu is outside the screen then GTK+ pushes it into the visible area. Otherwise the menu is cut of at the end of the visible screen area.

    Regardless of what happens to the size of the menu, the result is always that the items are placed in the same place as if the menu was placed outside the screen, using menu scrolling if necessary.

  • The "draw" signal and virtual method on GtkWidget has been removed. All drawing should now occur by invalidating a region of the widget (call gdk_window_invalidate_rect() or gtk_widget_queue_draw() for example to invalidate a region). GTK+ merges all invalid regions, and sends expose events to the widget in an idle handler for the invalid regions. gtk_widget_draw() is deprecated but still works; it adds the passed-in area to the invalid region and immediately sends expose events for the current invalid region. Most widgets will work fine if you just delete their "draw" implementation, since they will already have working expose_event implementations. The draw method was rarely called in practice anyway.

  • The GdkExposeEvent has a new region field. This can be used instead of the area field if you want a more exact representation of the area to update.

  • Sending synthetic exposes using gtk_widget_event() is no longer allowed. If you just need an expose call you should use gdk_window_invalidate_rect() or gdk_window_invalidate_region() instead. For the case of container widgets that need to propagate expose events to NO_WINDOW children you can either use gtk_container_propagate_expose(), or chain to the default container expose handler.

  • The draw_default and draw_focus methods/signals on GtkWidget are gone; simply draw things in your expose handler. gtk_widget_draw_focus() and gtk_widget_draw_default() wrapper functions are also gone; just queue a draw on the widget, or the part affected by the focus/default anyway. Also, GtkWidget now has default implementations for focus_in_event and focus_out_event. These set/unset GTK_HAS_FOCUS, and queue a draw. So if your focus in/out handler just does that, you can delete it.

  • GtkText and GtkTree are buggy and broken. We don't recommend using them, and changing old code to avoid them is a good idea. The recommended alternatives are GtkTextView and GtkTreeView. The broken widgets are not declared in the headers by default; to use them, define the symbol GTK_ENABLE_BROKEN during compilation. In some future release, these widgets will be removed from GTK+.

  • GdkColorContext is gone; you probably weren't using it anyway. Use GdkColormap and the gdk_rgb_* functions instead.

  • GtkMenuBar now draws the GtkContainer::border_width space outside the frame, not inside the frame.

  • In GTK+ 1.2, if an event handler returned TRUE it prevented propagation of that event to parent widgets. That is, the event signal would not be emitted on parent widgets. In GTK+ 2.0, if an event handler returns TRUE, the current signal emission on the current widget is immediately stopped. That is, other callbacks connected to the signal will not be invoked.

  • gtk_toolbar_new() no longer has arguments. This function was broken because the default GtkToolbarStyle (icons, text, both) is now a user preference, which is overridden when you call gtk_toolbar_set_style(). The constructor forced everyone to override the preference, which was undesirable. So to port your app, decide if you want to force the toolbar style or conform to the user's global defaults; if you want to force it, call gtk_toolbar_set_style().

    The orientation arg was removed from gtk_toolbar_new() as well, just because it wasn't very useful and we were breaking the function anyway so had an opportunity to lose it. Call gtk_toolbar_set_orientation() to set toolbar orientation.

  • GtkRange/GtkScrollbar/GtkScale were rewritten; this means that most theme engines won't draw them properly, and any custom subclasses of these widgets will need a rewrite (though if you could figure out how to subclass the old version of GtkRange, you have our respect). Also, GtkTroughType is gone.

    Here are some notable changes:

    • stepper_size style property is the height for vertical ranges, width for horizontal; the other dimension matches the trough size.

    • Added the ability to do NeXT-style steppers (and several other styles that don't make any sense).

    • Added min_slider_length, fixed_slider_length properties to GtkScrollbar.

    • Cleaned some private (or at least useless) functions out of gtkscale.h, e.g. gtk_scale_value_width.

    • Moved bindings from subclasses to GtkScale, even arrow keys, since blind users don't know scale orientation.

    • Changed move_slider action signal to use new GtkScrollType, remove GtkTroughType argument.

    • Digits rounds the values a range will input to the given number of decimals, but will not try to force adjustment values set by other controllers. That is, we no longer modify adjustment->value inside a value_changed handler.

    • Added getters for GtkScale setters.

    • Middle-click begins a slider drag.

  • The GtkContainer::focus signal/virtual function and gtk_container_focus() call were replaced by GtkWidget::focus and gtk_widget_child_focus(). The semantics are the same, so you should be able to just replace container_class->focus = mywidget_focus with widget_class->focus = mywidget_focus and replace gtk_container_focus() calls with gtk_widget_child_focus() calls.

    The purpose of this change was to allow non-containers to have focusable elements.

  • gtk_rc_set_image_loader() and gtk_rc_load_image() have been removed, now that GTK+ includes decent image loading capabilities itself.

  • An extra GtkSettings argument has been added to gtk_rc_find_pixmap_in_path(). This function is only actually useful from a theme engine during parsing, at which point the GtkSettings is provided.

  • The child argument facility in gtkcontainer.c has been converted to a child property facility using GParamSpec and other facilities for GObject.

  • The set_child_arg() and get_child_arg() virtual methods have been replaced with set_child_property()/get_child_property(), which work similar to GObject->set_property/get_property.

  • Other removed GtkContainer functions with the replacements:

    Old functionReplacement

  • gdk_image_get() (or rather its replacement, gdk_drawable_get_image()) now handles errors properly by returning NULL, previously it would crash. Also, a window being offscreen is no longer considered an error; instead, the area contains undefined contents for the offscreen areas. In most cases, code using gdk_image_get() should really be ported to gdk_pixbuf_get_from_drawable().

  • gtk_widget_set_usize() has been renamed to gtk_widget_set_size_request(), however the old name still exists unless you define GTK_DISABLE_DEPRECATED.

  • gtk_widget_set_uposition() is deprecated; use gtk_window_move(), gtk_fixed_put(), or gtk_layout_put() instead.

  • gtk_window_set_policy() is deprecated. To get the effect of "allow_shrink", call gtk_widget_set_size_request (window, 0, 0). To get the effect of "allow_grow", call gtk_window_set_resizable (window, TRUE). You didn't want the effect of "auto_shrink", it made no sense. But maybe if you were using it you want to use gtk_window_resize (window, 1, 1) to snap a window back to its minimum size (the 1, 1 will be rounded up to the minimum window size).

  • The core GTK+ now takes care of handling mapping, unmapping and realizing the child widgets of containers in gtk_widget_set_parent(). In most cases, this allows container implementations to be simplified by removing the code in add() methods to map and realize children. However, there are a couple of things to watch out for here:

  • If the parent is realized before the add() happens, gtk_widget_set_parent_window() must be called before gtk_widget_set_parent(), since gtk_widget_set_parent() will realize the child.

  • If a container depended on its children not being mapped unless it did so itself (for example, GtkNotebook only mapped the current page), then the new function gtk_widget_set_child_visible() must be called to keep widgets that should not be mapped not mapped.

    As part of this change, most containers also will no longer need custom implementations of the map() and unmap() virtual functions. The only cases where this is necessary are:

    • For !NO_WINDOW widgets, if you create children of widget->window and don't map them in realize() then you must map them in map(). [ In almost all cases, you can simply map the windows in realize(). ]

    • For NO_WINDOW widgets, if you create windows in your realize() method, you must map then in map() and unmap them in unmap().

  • gtk_widget_set_default_style(), gtk_widget_push_style(), and gtk_widget_pop_style() have been removed, since they did not work properly with themes and there were better alternatives for modifying the appearance of widgets. You should generally use gtk_widget_modify_*() instead.

  • gtk_image_new() now takes no arguments and creates an empty GtkImage widget. To create a GtkImage widget from a GdkImage (the least common usage of GdkImage), use gtk_image_new_from_image().

  • GTK_SELECTION_EXTENDED is now deprecated, and neither the GtkList/GtkTree nor the GtkCList/GtkCTree support GTK_SELECTION_EXTENDED anymore. However, the old extended behavior replaces MULTIPLE behavior.

  • The following variables are no longer exported from GDK. (Other variables are also no longer exported; the following are the ones found used externally in a large sample of GTK+ code.)

    gdk_null_window_warningsNone - did nothing in GTK+ 1.2
    gdk_leader_windowNone - private variable
    gdk_screengdk_x11_get_default_screen ()
    gdk_root_windowgdk_x11_get_default_root_xwindow ()
    gdk_root_parentgdk_get_default_root_window ()
    gdk_error_codegdk_error_trap_push ()/pop ()
    gdk_error_warningsgdk_error_trap_push ()/pop ()
    gdk_display_namegdk_get_display ()
    gdk_wm_delete_windowgdk_atom_intern ("WM_DELETE_WINDOW", FALSE)
    gdk_wm_take_focusgdk_atom_intern ("WM_TAKE_FOCUS", FALSE)
    gdk_wm_protocolsgdk_atom_intern ("WM_PROTOCOLS", FALSE)

  • The handling of colormaps and widgets has been changed:

    • The default colormap for widgets is now the GdkRGB colormap, not the system default colormap. If you try to use resources created for a widget (e.g., widget->style) with a window using the system colormap, errors will result on some machines.

    • gtk_widget_push()/gtk_widget_pop_colormap() only cause the colormap to be explicitly set on toplevel widgets, not on all widgets. The colormap for other widgets (when not set using gtk_widget_set_colormap()), is determined by finding the nearest ancestor with a colormap set on it explicitly, or if that fails, the default colormap.

  • The default selected day for GtkCalendar is now the current day in the month, not the first day in the month. The current month and year were already used.

  • GDK is no longer put into threaded mode automatically when g_thread_init() has been called. In order to use the global GDK thread mutex with gdk_threads_enter() and gdk_threads_leave(), you must call gdk_threads_init() explicitly. If you aren't using GDK and GTK+ functions from multiple threads, there is no reason to call gdk_threads_init().

  • The GtkPreviewInfo struct has had its visual and colormap fields removed. Also, gtk_preview_get_cmap() and gtk_preview_get_visual() are deprecated, as GdkRGB works on any colormap and visual. You no longer need to gtk_widget_push_cmap (gtk_preview_get_cmap ()) in your code.

  • The GtkBox, GtkTable, and GtkAlignment widgets now call gtk_widget_set_redraw_on_allocate (widget, FALSE); on themselves. If you want to actually draw contents in a widget derived from one of these widgets, you'll probably want to change this in your init() function.

  • A number of widgets are now NO_WINDOW widgets (most importantly GtkButton, but also GtkRange and GtkNotebook) This has a couple of effects:

    • If you are deriving from one of these widgets, you need to adapt your code appropriately -- for instance, drawing coordinates start from widget->allocation.x, widget->allocation.y.

    • If you are embedding one of these widgets in a custom widget, you must make sure you call gtk_container_propagate_expose() correctly, as you must for any NO_WINDOW widgets.

    GtkFixed is a little special; it is now created by default as a NO_WINDOW widget, but if you do

        gtk_fixed_set_has_window (fixed, TRUE);

    after creating a fixed widget, it will create a window and handle it properly.

  • GtkLayout no longer has the xoffset, yoffset fields, which used to store the difference between world and window coordinates for layout->bin_window. These coordinate systems are now always the same.

  • gtk_paint_focus(), gtk_draw_focus() and GtkStyle::draw_focus() have been changed a bit:

    • A GtkStateType argument has been added to gtk_paint_focus().

    • The default implementation of the GtkStyle::draw_focus() virtual function now draws a focus rectangle whose width is determined by the GtkWidget::focus-width style property.

    • The rectangle passed in is the bounding box, instead of the rectangle used in the gdk_draw_rectangle() call, so it is no longer necessary to subtract 1 from the width and height.