Weird bug with desynced nodes - including demo and screencast

Question

This is a follow-up to Desync of Elements and logic - how to debug?
I started a new thread because I will provide better information on this.
So we've made our graph editor working with JavaFX. One of our common demos for different products is a organizational chart demo. This is a tree-like graph with employees as nodes to depict the hierarchical structure of companies.
You can try this demo here (jre7u6 with JavaFX 2.2 b17 required):
http://www.yworks.com/products/yfilesfx/OrgChartDemo/OrgChartDemo.jnlp
We have also made a screencast which can be viewed here that shows the usage and most importantly the problem we are facing with JavaFX here:
http://www.screenr.com/uEF8
The startup is quite slow, it takes a few seconds for the graph to load. This is because the visual representation of the nodes is described within (nested) fxml files and each of those files is loaded in a not so effective way (because it is at the moment not possible in JavaFX to define a custom BeanAdapter in the FXMLLoader that does proper caching). After those files are loaded, there are still some ClassNotFoundExceptions thrown in the background by some part of the JavaFX framework I did not identify, yet. After a few seconds the performance speeds up significantly.
What you can do in this demo is described on the right. Basically you can zoom in/out, fit view, center a certain employee and interact with them.
Interaction means you can collapse and enlarge certain parts of the graph. For example, you can hide all subordinates of an employee and show them again.
The part of "showing them again" is done with an animation effect that is implemented in our framework, not in JavaFX.
When you hover an employee, a fading translation is played as a hover effect and some buttons appear. This part is important so i wanted to highlighted.
The hover effect is triggered by a onMouseEntered and onMouseExited event.
Also you can focus a certain employee by clicking on them. When clicked, the color changes from blue to orange. This is done by our internal logic, since the clickable things on the screen are not the JavaFX rectangles and such but rather a logical representation of a node with a position on the screen and a with and height.
So this is where things go wrong and what is really crucial for our application:
When you collapse/enlarge nodes (employees) a bug appears quite frequently that causes the internal representation and the actual appearence of JavaFX elements on the screen to be out of sync. Most notably the hover effect doesn't work anymore: The node is visually present (as you can see it) and the values of the nodes in the scenegraph match as well, but those nodes just are not present any more for any events. They do not receive mouse entered/exited events, hover-events and pressed-events and so on. You can still click on the node and focus it, but this is done in our framework (indicating that our logic works).
The connection of the logic and the javafx controller is done via databinding. The controller listens to changes in the corresponding model item and updates itself accordingly - you can see it when clicking on a node, the selection state of the model item changes and the controller updates the style of the javafx node.
The other effects however, like the hover-effect, are implemented on databinding on properties of the JavaFX node class: hoverProperty, onMouseEntered onMouseExited and so on.
Once a node is in this state, weird stuff is happening. For example, a node that is in this state seems to have normal values when viewed in ScenicView. However, when changing some values, for example translateX / Y, the node just disappears. (see also: http://www.screenr.com/lYF8 where i recorded that (a bit later in the video))
For an example of this have a look at the link to the screencast above. We showed how to reproduce the bug there.
We noticed that without any animation played this does not happen. So it seems to be caused by the animation. This is itself quite surprising, because we used the same animation process with other javaFX applications but never experienced this bug.
What happens is that our Animator sets the layoutX and layoutY values of a node that is being animated repeatedly and in rapid succession over a period of time.
The only thing we can think of that might cause this problem is that JavaFX somehow is not able to process a large amount of such update-calls and throws some exception internally and just dies.
Because of this we strongly hope that somebody can point us to were we have to look or how to even debug this problem.
* edit: added second screencast for ScenicView

Answer 1

jsmith wrote:
Just want to say thanks for the yworks stuff phelan.
I use the swing version embedded in Idea and the diagrams are really nice => everytime I make a printout I get asked "what tool did you use to generate that diagram?".:D thank you very much.
Only real advice I can offer is something you almost certainly know =>
JavaFX is a singlethreaded UI framework.
Ensure that anything you do which could read/write from nodes in an active scenegraph be done on the JavaFX application thread.I just double checked this and unless Platform.isFxApplicationThread() returns a wrong value, the only interaction with JavaFX nodes is done in the JavaFX application thread.
I think your app issues will prove very hard to resolve through this forum. Nicolas Lorain is the javafx product manager, so you could try contacting him directly for support. There is also some kind of JavaFX partner program which may be able to offer support as well, Nicolas should also have details on that.
Good luck with your app.I just searched for the JavaFX partner program, and couldn't find anything ongoing. The only advanced support I could find is the Oracle Premier Support, which at this point isn't suitable for us. I also couldn't find any contact information for Nicolas Lorain (he has a blog and twitter but I cannot find his email). So if you could point me to somewhere, I'd be very glad.