]> GTK+ 3 is a major new version of GTK+ that breaks both API and ABI compared to GTK+ 2.x, which has remained API- and ABI-stable for a long time. Thankfully, most of the changes are not hard to adapt to and there are a number of steps that you can take to prepare your GTK+ 2.x application for the switch to GTK+ 3. After that, there's a small number of adjustments that you may have to do when you actually switch your application to build against GTK+ 3.

The steps outlined in the following sections assume that your application is working with GTK+ 2.24, which is the final stable release of GTK+ 2.x. It includes all the necessary APIs and tools to help you port your application to GTK+ 3. If you are still using an older version of GTK+ 2.x, you should first get your application to build and work with 2.24.

With GTK+ 2.x it was common to include just the header files for a few widgets that your application was using, which could lead to problems with missing definitions, etc. GTK+ 3 tightens the rules about which header files you are allowed to include directly. The allowed header files are are gtk/gtk.h for GTK gtk/gtkunixprint.h for low-level, UNIX-specific printing functions gdk/gdk.h for GDK gdk/gdkx.h for GDK functions that are X11-specific gdk/gdkwin32.h for GDK functions that are Windows-specific (these relative paths are assuming that you are using the include paths that are specified in the gtk+-2.0.pc file, as returned by pkg-config --cflags gtk+-2.0.pc.) To check that your application only includes the allowed headers, you can use defines to disable inclusion of individual headers, as follows: make CFLAGS+="-DGTK_DISABLE_SINGLE_INCLUDES"

Over the years, a number of functions, and in some cases, entire widgets have been deprecated. These deprecations are clearly spelled out in the API reference, with hints about the recommended replacements. The API reference also includes an index of all deprecated symbols. To verify that your program does not use any deprecated symbols, you can use defines to remove deprecated symbols from the header files, as follows: make CFLAGS+="-DGDK_DISABLE_DEPRECATED -DGTK_DISABLE_DEPRECATED"

GTK+ 3 removes many implementation details and struct members from its public headers. To ensure that your application does not have problems with this, you define the preprocessor symbol GSEAL_ENABLE while building your application against GTK+ 2.x. This will make the compiler catch all uses of direct access to struct fields so that you can go through them one by one and replace them with a call to an accessor function instead. make CFLAGS+="-DGSEAL_ENABLE"

Key constants have gained a _KEY_ infix. For example, GDK_a is now GDK_KEY_a. In GTK+ 2, the old names continue to be available. In GTK+ 3 however, the old names will require an explicit include of the gdkkeysyms-compat.h header.

The gdk_spawn family of functions has been deprecated in GDK 2.24 and removed from GDK 3. Various replacements exist; the best replacement depends on the circumstances: If you are opening a document or URI by launching a command like firefox http://my-favourite-website.com or gnome-open ghelp:epiphany, it is best to just use gtk_show_uri(); as an added benefit, your application will henceforth respect the users preference for what application to use. If you are launching a regular, installed application that has a desktop file, it is best to use GIOs #GAppInfo with a suitable launch context. GAppInfo *info; GAppLaunchContext *context; GError *error = NULL; info = g_desktop_app_info_new ("epiphany.desktop"); context = gdk_display_get_app_launch_context (display); g_app_info_launch (info, NULL, context, &error); if (error) { g_warning ("Failed to launch epiphany: %s", error->message); g_error_free (error); } g_object_unref (info); g_object_unref (context); If you are launching a custom commandline, you can still use g_app_info_launch() with a GAppInfo that is constructed with g_app_info_create_from_commandline(), or you can use the more lowlevel g_spawn family of functions (e.g. g_spawn_command_line_async()), and pass DISPLAY in the environment. gdk_screen_make_display_name() can be used to find the right value for the DISPLAY environment variable.

In GTK+ 3, the GDK drawing API (which closely mimics the X drawing API, which is itself modeled after PostScript) has been removed. All drawing in GTK+ 3 is done via cairo. The #GdkGC and #GdkImage objects, as well as all the functions using them, are gone. This includes the gdk_draw family of functions like gdk_draw_rectangle() and gdk_draw_drawable(). As #GdkGC is roughly equivalent to #cairo_t and #GdkImage was used for drawing images to GdkWindows, which cairo supports automatically, a transition is usually straightforward. The following examples show a few common drawing idioms used by applications that have been ported to use cairo and how the code was replaced. Drawing a pixbuf onto a drawable used to be done like this: black_gc, pixbuf, 0, 0 x, y, gdk_pixbuf_get_width (pixbuf), gdk_pixbuf_get_height (pixbuf), GDK_RGB_DITHER_NORMAL, 0, 0); ]]> Doing the same thing with cairo: Note that very similar code can be used for drawing pixmaps by using gdk_cairo_set_source_pixmap() instead of gdk_cairo_set_source_pixbuf(). Tiled pixmaps are often used for drawing backgrounds. Old code looked something like this: black_gc; gdk_gc_set_tile (gc, pixmap); gdk_gc_set_fill (gc, GDK_TILED); gdk_gc_set_ts_origin (gc, x_origin, y_origin); /* use */ gdk_draw_rectangle (drawable, gc, TRUE, 0, 0, width, height); /* restore */ gdk_gc_set_tile (gc, NULL); gdk_gc_set_fill (gc, GDK_SOLID); gdk_gc_set_ts_origin (gc, 0, 0); ]]> The equivalent cairo code looks like this: Again, you can exchange pixbufs and pixmaps by using gdk_cairo_set_source_pixbuf() instead of gdk_cairo_set_source_pixmap(). Drawing layouts clipped is often used to avoid overdraw or to allow drawing selections. Code would have looked like this: text_gc[state]; gdk_gc_set_clip_rectangle (gc, &area); /* use */ gdk_draw_layout (drawable, gc, x, y, layout); /* restore */ gdk_gc_set_clip_rectangle (gc, NULL); ]]> With cairo, the same effect can be achieved using: text[state]); /* draw the text */ cairo_move_to (cr, x, y); pango_cairo_show_layout (cr, layout); cairo_destroy (cr); ]]> Clipping using cairo_clip() is of course not restricted to text rendering and can be used everywhere where GC clips were used. And using gdk_cairo_set_source_color() with style colors should be used in all the places where a style’s GC was used to achieve a particular color.

Stippling is the usage of a bi-level mask, called a #GdkBitmap. It was often used to achieve a checkerboard effect. You can use cairo_mask() to achieve this effect. To get a checkerbox mask, you can use code like this: Note that stippling looks very outdated in UIs, and is rarely used in modern applications. All properties that made use of stippling have been removed from GTK+ 3. Most prominently, stippling is absent from text rendering, in particular #GtkTextTag. The gdk_draw_drawable() function allowed using the same drawable as source and target. This was often used to achieve a scrolling effect. Cairo does not allow this yet. You can however use cairo_push_group() to get a different intermediate target that you can copy to. So you can replace this code: By using this code: The cairo developers plan to add self-copies in the future to allow exactly this effect, so you might want to keep up on cairo development to be able to change your code. GDK provided a way to ignore the color attributes of text and use a hardcoded text color with the gdk_draw_layout_with_colors() function. This is often used to draw text shadows or selections. Pango’s cairo support does not yet provide this functionality. If you use Pango layouts that change colors, the easiest way to achieve a similar effect is using pango_cairo_layout_path() and cairo_fill() instead of gdk_draw_layout_with_colors(). Note that this results in a slightly uglier-looking text, as subpixel anti-aliasing is not supported.

This section outlines porting tasks that you need to tackle when you get to the point that you actually build your application against GTK+ 3. Making it possible to prepare for these in GTK+ 2.24 would have been either impossible or impractical.

The request-phase of the traditional GTK+ geometry management has been replaced by a more flexible height-for-width system, which is described in detail in the API documentation (see ). As a consequence, the ::size-request signal and vfunc has been removed from #GtkWidgetClass. The replacement for size_request() can take several levels of sophistication: As a minimal replacement to keep current functionality, you can simply implement the #GtkWidgetClass.get_preferred_width() and #GtkWidgetClass.get_preferred_height() vfuncs by calling your existing size_request() function. So you go from static void my_widget_class_init (MyWidgetClass *class) { GtkWidgetClass *widget_class = GTK_WIDGET_CLASS (class); /* ... */ widget_class->size_request = my_widget_size_request; /* ... */ } to something that looks more like this: static void my_widget_get_preferred_width (GtkWidget *widget, gint *minimal_width, gint *natural_width) { GtkRequisition requisition; my_widget_size_request (widget, &requisition); *minimal_width = *natural_width = requisition.width; } static void my_widget_get_preferred_height (GtkWidget *widget, gint *minimal_height, gint *natural_height) { GtkRequisition requisition; my_widget_size_request (widget, &requisition); *minimal_height = *natural_height = requisition.height; } /* ... */ static void my_widget_class_init (MyWidgetClass *class) { GtkWidgetClass *widget_class = GTK_WIDGET_CLASS (class); /* ... */ widget_class->get_preferred_width = my_widget_get_preferred_width; widget_class->get_preferred_height = my_widget_get_preferred_height; /* ... */ } Sometimes you can make things a little more streamlined by replacing your existing size_request() implementation by one that takes an orientation parameter: static void my_widget_get_preferred_size (GtkWidget *widget, GtkOrientation orientation, gint *minimal_size, gint *natural_size) { /* do things that are common for both orientations ... */ if (orientation == GTK_ORIENTATION_HORIZONTAL) { /* do stuff that only applies to width... */ *minimal_size = *natural_size = ... } else { /* do stuff that only applies to height... */ *minimal_size = *natural_size = ... } } static void my_widget_get_preferred_width (GtkWidget *widget, gint *minimal_width, gint *natural_width) { my_widget_get_preferred_size (widget, GTK_ORIENTATION_HORIZONTAL, minimal_width, natural_width); } static void my_widget_get_preferred_height (GtkWidget *widget, gint *minimal_height, gint *natural_height) { my_widget_get_preferred_size (widget, GTK_ORIENTATION_VERTICAL, minimal_height, natural_height); } /* ... */ If your widget can cope with a small size, but would appreciate getting some more space (a common example would be that it contains ellipsizable labels), you can do that by making your #GtkWidgetClass.get_preferred_width() / #GtkWidgetClass.get_preferred_height() functions return a smaller value for @minimal than for @natural. For @minimal, you probably want to return the same value that your size_request() function returned before (since size_request() was defined as returning the minimal size a widget can work with). A simple way to obtain good values for @natural, in the case of containers, is to use gtk_widget_get_preferred_width() and gtk_widget_get_preferred_height() on the children of the container, as in the following example: static void gtk_fixed_get_preferred_height (GtkWidget *widget, gint *minimum, gint *natural) { GtkFixed *fixed = GTK_FIXED (widget); GtkFixedPrivate *priv = fixed->priv; GtkFixedChild *child; GList *children; gint child_min, child_nat; *minimum = 0; *natural = 0; for (children = priv->children; children; children = children->next) { child = children->data; if (!gtk_widget_get_visible (child->widget)) continue; gtk_widget_get_preferred_height (child->widget, &child_min, &child_nat); *minimum = MAX (*minimum, child->y + child_min); *natural = MAX (*natural, child->y + child_nat); } } Note that the #GtkWidgetClass.get_preferred_width() / #GtkWidgetClass.get_preferred_height() functions only allow you to deal with one dimension at a time. If your size_request() handler is doing things that involve both width and height at the same time (e.g. limiting the aspect ratio), you will have to implement #GtkWidgetClass.get_preferred_height_for_width() and #GtkWidgetClass.get_preferred_width_for_height(). To make full use of the new capabilities of the height-for-width geometry management, you need to additionally implement the #GtkWidgetClass.get_preferred_height_for_width() and #GtkWidgetClass.get_preferred_width_for_height(). For details on these functions, see .

Starting with version 1.10, cairo provides a region API that is equivalent to the GDK region API (which was itself copied from the X server). Therefore, the region API has been removed in GTK+ 3. Porting your application to the cairo region API should be a straight find-and-replace task. Please refer to the following table: GDKcairo#GdkRegion#cairo_region_t#GdkRectangle#cairo_rectangle_int_tgdk_rectangle_intersect()this function is still theregdk_rectangle_union()this function is still theregdk_region_new()cairo_region_create()gdk_region_copy()cairo_region_copy()gdk_region_destroy()cairo_region_destroy()gdk_region_rectangle()cairo_region_create_rectangle()gdk_region_get_clipbox()cairo_region_get_extents()gdk_region_get_rectangles()cairo_region_num_rectangles() and cairo_region_get_rectangle()gdk_region_empty()cairo_region_is_empty()gdk_region_equal()cairo_region_equal()gdk_region_point_in()cairo_region_contains_point()gdk_region_rect_in()cairo_region_contains_rectangle()gdk_region_offset()cairo_region_translate()gdk_region_union_with_rect()cairo_region_union_rectangle()gdk_region_intersect()cairo_region_intersect()gdk_region_union()cairo_region_union()gdk_region_subtract()cairo_region_subtract()gdk_region_xor()cairo_region_xor()gdk_region_shrink()no replacementgdk_region_polygon()no replacement, use cairo paths instead

The #GdkPixmap object and related functions have been removed. In the cairo-centric world of GTK+ 3, cairo surfaces take over the role of pixmaps. One place where pixmaps were commonly used is to create custom cursors: GdkCursor *cursor; GdkPixmap *pixmap; cairo_t *cr; GdkColor fg = { 0, 0, 0, 0 }; pixmap = gdk_pixmap_new (NULL, 1, 1, 1); cr = gdk_cairo_create (pixmap); cairo_rectangle (cr, 0, 0, 1, 1); cairo_fill (cr); cairo_destroy (cr); cursor = gdk_cursor_new_from_pixmap (pixmap, pixmap, &fg, &fg, 0, 0); g_object_unref (pixmap); The same can be achieved without pixmaps, by drawing onto an image surface: GdkCursor *cursor; cairo_surface_t *s; cairo_t *cr; GdkPixbuf *pixbuf; s = cairo_image_surface_create (CAIRO_FORMAT_A1, 3, 3); cr = cairo_create (s); cairo_arc (cr, 1.5, 1.5, 1.5, 0, 2 * M_PI); cairo_fill (cr); cairo_destroy (cr); pixbuf = gdk_pixbuf_get_from_surface (NULL, s, 0, 0, 0, 0, 3, 3); cairo_surface_destroy (s); cursor = gdk_cursor_new_from_pixbuf (display, pixbuf, 0, 0); g_object_unref (pixbuf);

For drawing with cairo, it is not necessary to allocate colors, and a #GdkVisual provides enough information for cairo to handle colors in 'native' surfaces. Therefore, #GdkColormap and related functions have been removed in GTK+ 3, and visuals are used instead. The colormap-handling functions of #GtkWidget (gtk_widget_set_colormap(), etc) have been removed and gtk_widget_set_visual() has been added. You might have a screen-changed handler like the following to set up a translucent window with an alpha-channel: static void on_alpha_screen_changed (GtkWidget *widget, GdkScreen *old_screen, GtkWidget *label) { GdkScreen *screen = gtk_widget_get_screen (widget); GdkColormap *colormap = gdk_screen_get_rgba_colormap (screen); if (colormap == NULL) colormap = gdk_screen_get_default_colormap (screen); gtk_widget_set_colormap (widget, colormap); } With visuals instead of colormaps, this will look as follows: static void on_alpha_screen_changed (GtkWindow *window, GdkScreen *old_screen, GtkWidget *label) { GdkScreen *screen = gtk_widget_get_screen (GTK_WIDGET (window)); GdkVisual *visual = gdk_screen_get_rgba_visual (screen); if (visual == NULL) visual = gdk_screen_get_system_visual (screen); gtk_widget_set_visual (window, visual); }

In GTK+ 2.x, GDK could only be compiled for one backend at a time, and the %GDK_WINDOWING_X11 or %GDK_WINDOWING_WIN32 macros could be used to find out which one you are dealing with: #ifdef GDK_WINDOWING_X11 if (timestamp != GDK_CURRENT_TIME) gdk_x11_window_set_user_time (gdk_window, timestamp); #endif #ifdef GDK_WINDOWING_WIN32 /* ... win32 specific code ... */ #endif In GTK+ 3, GDK can be built with multiple backends, and currently used backend has to be determined at runtime, typically using type-check macros on a #GdkDisplay or #GdkWindow. You still need to use the #GDK_WINDOWING macros to only compile code referring to supported backends: #ifdef GDK_WINDOWING_X11 if (GDK_IS_X11_DISPLAY (display)) { if (timestamp != GDK_CURRENT_TIME) gdk_x11_window_set_user_time (gdk_window, timestamp); } else #endif #ifdef GDK_WINDOWING_WIN32 if (GDK_IS_WIN32_DISPLAY (display)) { /* ... win32 specific code ... */ } else #endif { g_warning ("Unsupported GDK backend"); }

The GtkWidget #GtkWidget::expose-event signal has been replaced by a new #GtkWidget::draw signal, which takes a #cairo_t instead of an expose event. The cairo context is being set up so that the origin at (0, 0) coincides with the upper left corner of the widget, and is properly clipped. In other words, the cairo context of the draw signal is set up in 'widget coordinates', which is different from traditional expose event handlers, which always assume 'window coordinates'. The widget is expected to draw itself with its allocated size, which is available via the new gtk_widget_get_allocated_width() and gtk_widget_get_allocated_height() functions. It is not necessary to check for GTK_WIDGET_IS_DRAWABLE(), since GTK+ already does this check before emitting the ::draw signal. There are some special considerations for widgets with multiple windows. Expose events are window-specific, and widgets with multiple windows could expect to get an expose event for each window that needs to be redrawn. Therefore, multi-window expose event handlers typically look like this: if (event->window == widget->window1) { /* ... draw window1 ... */ } else if (event->window == widget->window2) { /* ... draw window2 ... */ } ... In contrast, the ::draw signal handler may have to draw multiple windows in one call. GTK+ has a convenience function gtk_cairo_should_draw_window() that can be used to find out if a window needs to be drawn. With that, the example above would look like this (note that the 'else' is gone): if (gtk_cairo_should_draw_window (cr, widget->window1) { /* ... draw window1 ... */ } if (gtk_cairo_should_draw_window (cr, widget->window2) { /* ... draw window2 ... */ } ... Another convenience function that can help when implementing ::draw for multi-window widgets is gtk_cairo_transform_to_window(), which transforms a cairo context from widget-relative coordinates to window-relative coordinates. All GtkStyle drawing functions (gtk_paint_box(), etc) have been changed to take a #cairo_t instead of a window and a clip area. ::draw implementations will usually just use the cairo context that has been passed in for this. gboolean gtk_arrow_draw (GtkWidget *widget, cairo_t *cr) { gint x, y; gint width, height; gint extent; width = gtk_widget_get_allocated_width (widget); height = gtk_widget_get_allocated_height (widget); extent = MIN (width - 2 * PAD, height - 2 * PAD); x = PAD; y = PAD; gtk_paint_arrow (gtk_widget_get_style (widget), cr, gtk_widget_get_state (widget), GTK_SHADOW_OUT, widget, "arrow", widget->priv->arrow_type, TRUE, x, y, extent, extent); }

In GTK+ 2.x, #GtkProgressBar and #GtkCellRendererProgress were using the GtkProgressBarOrientation enumeration to specify their orientation and direction. In GTK+ 3, both the widget and the cell renderer implement #GtkOrientable, and have an additional 'inverted' property to determine their direction. Therefore, a call to gtk_progress_bar_set_orientation() needs to be replaced by a pair of calls to gtk_orientable_set_orientation() and gtk_progress_bar_set_inverted(). The following values correspond: GTK+ 2.xGTK+ 3GtkProgressBarOrientationGtkOrientationinvertedGTK_PROGRESS_LEFT_TO_RIGHTGTK_ORIENTATION_HORIZONTALFALSEGTK_PROGRESS_RIGHT_TO_LEFTGTK_ORIENTATION_HORIZONTALTRUEGTK_PROGRESS_TOP_TO_BOTTOMGTK_ORIENTATION_VERTICALFALSEGTK_PROGRESS_BOTTOM_TO_TOPGTK_ORIENTATION_VERTICALTRUE

The behaviour of expanding widgets has changed slightly in GTK+ 3, compared to GTK+ 2.x. It is now 'inherited', i.e. a container that has an expanding child is considered expanding itself. This is often the desired behaviour. In places where you don't want this to happen, setting the container explicity as not expanding will stop the expand flag of the child from being inherited. See gtk_widget_set_hexpand() and gtk_widget_set_vexpand().

The default values for the #GtkScrolledWindow:hscrollbar-policy and #GtkScrolledWindow:vscrollbar-policy properties have been changed from 'never' to 'automatic'. If your application was relying on the default value, you will have explicitly set it explicitly. The ::set-scroll-adjustments signal on GtkWidget has been replaced by the #GtkScrollable interface which must be implemented by a widget that wants to be placed in a #GtkScrolledWindow. Instead of emitting ::set-scroll-adjustments, the scrolled window simply sets the #GtkScrollable::hadjustment and #GtkScrollable::vadjustment properties.

GtkObject has been removed in GTK+ 3. Its remaining functionality, the ::destroy signal, has been moved to GtkWidget. If you have non-widget classes that are directly derived from GtkObject, you have to make them derive from #GInitiallyUnowned (or, if you don't need the floating functionality, #GObject). If you have widgets that override the destroy class handler, you have to adust your class_init function, since destroy is now a member of GtkWidgetClass: GtkObjectClass *object_class = GTK_OBJECT_CLASS (class); object_class->destroy = my_destroy; becomes GtkWidgetClass *widget_class = GTK_WIDGET_CLASS (class); widget_class->destroy = my_destroy; In the unlikely case that you have a non-widget class that is derived from GtkObject and makes use of the destroy functionality, you have to implement ::destroy yourself.

The resize grip functionality has been moved from #GtkStatusbar to #GtkWindow. Any window can now have resize grips, regardless whether it has a statusbar or not. The functions gtk_statusbar_set_has_resize_grip() and gtk_statusbar_get_has_resize_grip() have disappeared, and instead there are now gtk_window_set_has_resize_grip() and gtk_window_get_has_resize_grip().

Linking against GTK+ 2.x and GTK+ 3 in the same process is problematic and can lead to hard-to-diagnose crashes. The gtk_init() function in both GTK+ 2.22 and in GTK+ 3 tries to detect this situation and abort with a diagnostic message, but this check is not 100% reliable (e.g. if the problematic linking happens only in loadable modules). Direct linking of your application against both versions of GTK+ is easy to avoid; the problem gets harder when your application is using libraries that are themselves linked against some version of GTK+. In that case, you have to verify that you are using a version of the library that is linked against GTK+ 3. If you are using packages provided by a distributor, it is likely that parallel installable versions of the library exist for GTK+ 2.x and GTK+ 3, e.g for vte, check for vte3; for webkitgtk look for webkitgtk3, and so on.

Some software packages install loadable GTK+ modules such as theme engines, gdk-pixbuf loaders or input methods. Since GTK+ 3 is parallel-installable with GTK+ 2.x, the two GTK+ versions have separate locations for their loadable modules. The location for GTK+ 2.x is libdir/gtk-2.0 (and its subdirectories), for GTK+ 3 the location is libdir/gtk-3.0 (and its subdirectories). For some kinds of modules, namely input methods and pixbuf loaders, GTK+ keeps a cache file with extra information about the modules. For GTK+ 2.x, these cache files are located in sysconfdir/gtk-2.0. For GTK+ 3, they have been moved to libdir/gtk-3.0/3.0.0/. The commands that create these cache files have been renamed with a -3 suffix to make them parallel-installable. Note that GTK+ modules often link against libgtk, libgdk-pixbuf, etc. If that is the case for your module, you have to be careful to link the GTK+ 2.x version of your module against the 2.x version of the libraries, and the GTK+ 3 version against hte 3.x versions. Loading a module linked against libgtk 2.x into an application using GTK+ 3 will lead to unhappiness and must be avoided.