One of the questions asked in my JavaFX 2.0 blog post, and also at my second JavaFX 2.0 talk on Thursday at JavaOne was around Swing integration for JavaFX 2.0. I wanted to clarify what my current thinking is, and give you a chance to respond.
Edit: If you haven’t yet, please go read Amy’s heartfelt ramble on swing and javafx. She’s says better (and with more authority, having been a critical engineer on AWT, Swing, and JavaFX) than I did.
Over the past several years as I’ve gone on customer visits or to conferences such as Devoxx, I’ve heard repeatedly and clearly that people want to be able to use JavaFX in their existing swing applications. This is obviously important for anybody who has a big, existing swing application or is planning on building a new application in Java and has chosen swing.
In the JavaFX 1.x platform line, we’ve supported the ability to embed Swing components in JavaFX applications. This, however, is not generally that useful for people with existing Swing applications, but rather for people writing new JavaFX applications. This ability has been crucial if they wanted to have rich text or tables or combo boxes. Basically, it acted as a patch to allow JavaFX apps to be built if there weren’t equivalent JavaFX UI Controls yet built. But it wasn’t a bridge per se.
It also isn’t supported in the new Prism pipeline.
However going the other way and allowing JavaFX to be embedded in Swing should be quite feasible, even for the Prism pipeline. Which gets to another point — why having Java APIs for JavaFX was so important.
While on the train to the first Devoxx after JavaFX 1.0 was released, Jasper and I whacked together a quick proof-of-concept implementation of a Swing component that could host a JavaFX Scene. It was subsequently released by Josh Marinacci and some other implementation (I’m not sure if it was ultimately based on what Jasper and I did) was taken and released as part of JFXtras.
The problem with this has been that you cannot reliably modify the JavaFX scene from Java code. This is because the JavaFX Script compiler would produce “obfuscated” java code (this is actually a huge oversimplification and not quite technically accurate, but correct in the essentials). Instead of the mutator for the “x” property being setX, it was set$x. But of course what it was actually called and how it actually worked was subject to change from release to release so you couldn’t dependably talk from Java to JavaFX Script except through interfaces — it was all quite cumbersome.
So to enable developers to embed JavaFX inside Swing is a two part problem: how to do it at a graphics engine level, and provide clean Java APIs for working with JavaFX.
Now that we’ve announced the latter, we’re ready to start working on the former. There are, as you can imagine, a lot of details to sort through. For example, do we force 2D rendering when embedding in Swing? That would certainly be the easiest thing for us to do, but also the least glamourous since you wouldn’t get all that nifty hardware accelerated 3D stuff we’re talking about.
We may instead try to do something along the lines of what Java3D and JOGL did — essentially embed a heavyweight in the swing app. We’ll have to work through problems that may arise when Java2D is using DirectX and JavaFX is using DirectX. We’ll also have a whole pile of thread related issues to sort through.
Speaking of threads, one of the main design goals for JavaFX was to separate the UI thread from the rendering threads. This would allow us to take advantage of all those wonderful cores on machines these days for rendering without slowing down event response times etc. However, Swing was designed such that the UI thread *was* the rendering thread (another problem with immediate mode rendering APIs).
I guess I haven’t quite recovered from JavaOne yet. Rereading this blog post it rambles a bit. So I’d better just stop :-). In Summary, the current plan of record is to work towards allowing developers to put JavaFX content into Swing applications (WebView, Media, scenes, animations, etc), but perhaps not the other way around.