Archive for March, 2008|Monthly archive page

GTK+ 3.0: Getting serious.

GTK+ has come a long way. From its humble beginnings as “The GIMP ToolKit”, it is now used in a plethora of applications. In fact, GTK+ is very popular. GNOME, one of the leading desktop environment on Unix systems, uses GTK+ almost exclusively. The Gimp is built upon GTK+, of course. And there are many commercial software developers like Adobe, NVidia and VMware that decided to use it as a base for their products.

Still, there are several shortcomings with GTK+. Development of the 2.x series started back in 2002. Since then, GTK+ has ripened and aged. It has aged well, but still: its age shows.  Throughout the 2.x cycle, several years now, the developers have kept GTK+ ABI compatible. This keeps application developers depending on GTK+ very happy:  they can be sure that code linked to an older version of GTK+ continues to work with newer releases. Packages released back in 2002 will continue to work with new library releases. That’s great, because no third-party application developer likes to rebuild and repackage the whole product line, just because a new version of the underlying libraries got released and all the distribution start packaging that version. It causes work and trouble. And for commercial/proprietary developers that means costs.

The commitment not to break ABI made a lot of people very happy. But it also put very tight constraints on the GTK+ developers. It’s not that easy to add new features and remain ABI compatible. Minor features, yes. But as soon as you want to make radical improvements and need to change the exposed data structures, you run into serious trouble. It just not possible beyond a certain point.

On the 2008 GTK+ Hackfest in Berlin, Imendio’s GTK+ hackers presented their vision [1] of GTK+’s future and the reasons why they think that GTK+ has to make a step forward, embrace change and break ABI compatibility. Other GTK+ developers [2] have also voiced their opinions, listing parts of GTK+ that need serious love, but state that they don’t require breakage.
Whether or not these are the things that will mark the road to GTK+ 3.0, almost all of them need attention. And give hints to the shape of things to come.

Theming is one major aspect of GTK+ that needs a serious overhaul. Theming in GTK+ sucks and blows big time. The initial concept of how theming works in GTK+ stems from the very first releases and never received serious love. As a result it is very difficult to do fancy graphic things in GTK+ or to make custom widgets that fit into the rest of the desktop. The funny look of Evolution’s tree headers in some themes is one symptom, but every developer with the need to write custom widgets is looking for a hard time.
There have been several suggestions on how to do that, some of them involving CSS-style theming [3]. CSS would be nice, for sure. But even the ability to paint one widget to mimick another would be a huge gain. Application-specific theming and custom layouts? Delicious.

Although it is possible to create animations in plain GTK+, it’s not very easy to do. Out of the desire to create fancy interfaces in the image of the iPhone interface arose several GLib/GTK+ inspired libraries; Clutter, Pigment and Moonlight. All of those have drawbacks, however: Clutter doesn’t use the GLib event system, Moonlight is written in C++ (a no-go for a GTK+ library) and Pigment is in a very rough state. Still, there are very solid plans to what extent a scene-graph library might interact with GTK+ and what requirements such a library has to fulfill [4].

GTK+ has no standard Canvas. There is a GnomeCanvas, but it’s deprecated, not very popular and lacks some key features, like drawing GUI. Many developers resort to plain Cairo when it comes to custom graphics, but Cairo is lacking a way to draw GUI elements also. Nothing gained. There are some possible candidates [5] for a possible GTKCanvas, but none of them seems to be the right candidate. And then there is the question, if a specialised canvas is a good idea at all.
This problem might be solved with the emergence of the aforementioned scene-graph library; instead of introducing a specialised library for custom paint operations, make that library the standard way.

OS integration:
GTK+ is not limited to X11 systems anymore. There are many GTK+ applications that have been ported to Windows and enjoy a surprising popularity there; Inkscape for example has a significant Windows user base. And OS X gets more important with every passing month. Some of these applications make extensive use of operating system features. Up to now, GTK+ featured only a limited set of functions to provide access to operating system functions, but the first solutions addressing this problem are starting to appear [6].

One of the GTK+ buzzwords of the last few month has been introspection. Introspection allows to, well, inspect an object, its methods, public members and its inheritance. This is not only very comfortable for debugging, it also allows for very easy bindings: automated bindings for your favourite programming language? Here it comes. It might still be a while until all parts are in place, but already the results are amazing [7].

It might still be a long time until GTK+ 3.0 gets released. And in any way; GTK+ 3.0 won’t be about adding new features. There are still some mistakes of the past lumbering in GTK+. Exposed private structues, public members that get manipulated directly: things like these have to be fixed before a GTK+ 3.2 can start adding features [8].  But with some of the features, especially a scene-graph, window-independent widget placing and over-rideable paint methods for GTK+ widgets, GTK+ is starting to look very interesting again.

[1] Imendio’s GTK+ 3.0 vision
[2] Gtk+ Hackfest 2008: Day one and a half
[3] GUADEC 5 Notes
[4] New Theme API
[5] Canvas Overview
[6] libgtkhotkey 0.1, Portable Hotkey Integration
Future of GNOME language bindings

[8] GTK+ 3.0: Enabling incrementalism