Thanks, it seems I misunderstood something fundamental here.
I always figured the memory effects are limited to the variables
accessed/mutated within a synchronized block (or *could* be limited to
just those by selectively flushing data by some advanced VM).
Apparently that's not the case, and a full flush is simply done after a
synchronized block completes, which would include all mutated fields,
not just those inside a synchronized block.
--John
On 25/01/2024 15:50, Kevin Rushforth wrote:
Yes, I think you are missing something. Unless you have an animation
that just decides all on it's own to find and attach a newly
discovered node to the scene and start rendering it, there will be a
synchronization as a result of calling runLater. Consider the
following code:
Label label = new Label();
label.setText("hi");
// Attach it to the scene on the FX thread.
Platform.runLater(() -> root.getChildren().add(label));
The "runLater" necessarily synchronizes with the FX Application thread
so that the runnable can be put on its queue and later taken off the
queue and executed on the FX Application thread. As specified in the
Platform::runLater API docs: Actions in a thread prior to submitting a
runnable to this method happen-before actions performed by the
runnable in the JavaFX Application Thread.
So I don't see the problem.
-- Kevin
On 1/25/2024 1:50 AM, John Hendrikx wrote:
All this threading talk has made me wonder something:
Let's say there are two threads, the FX thread and Thread 1. I do
the following:
- On thread 1: create Label
- On thread 1: set Label text to "xyz"
I now attach this label to an active Scene graph. This should be done
on the FX thread as we're going to be manipulating an active Scene
graph, so:
- On FX thread: attach Label to active scene graph
There is no synchronization in place on the Label's text field. What
guarantees do I have that when the label is shown on screen it will
show "xyz" ?
IMHO, there is no such guarantee, and so any creation or manipulation
of Nodes that will later be part of an active scene graph (and thus
accessed by the FX thread) **must** be done on the FX thread.
Involving any other thread in their creation or manipulation runs the
risk of seeing an object in the incorrect state (or even an
"impossible" state when multiple fields are involved that normally
change together).
Effectively, assuming that when you create Nodes you always have the
intention of showing them at some point, you can never construct
Nodes on any other thread than the FX thread...
Am I missing something?
--John
