fair-acc/chart-fx integration

[flef]

Hello,

I heard about a very interesting project for high-frequency data analysis and visualization: fair-acc/chart-fx. Have you heard about this project?

Do you think integration into Phoebus is possible? DataBrowser already offers many possibilities, but this module goes even further.

https://github.com/fair-acc/chart-fx

Regards,
Florian.

[kasemir]

Plotting libraries come and go. This one indeed looks very good!

Most important, the basic DataSet is an interface, basically

public interface DataSet 
{
    int getDataCount();
    double get(final int dimIndex, final int index);

(https://github.com/fair-acc/chart-fx/blob/main/chartfx-dataset/src/main/java/io/fair_acc/dataset/DataSet.java)

I like that very much, because our data interface is exactly that.
The data browser keeps the data for a trace as a combination of ring buffer for recent live samples and an array for the historic samples. We can provide the number of samples and also an indexed sample.
Other plotting libs want you to provide "double[] x", so for each plot update we'd have to copy the data from the ring buffer and the historic samples into a new array.

One more thing: In the data browser, the data can change at any time as you receive updates from CA or PVA, and the plot can also be updated at any time, so our data has a Lock. The chartfx DataSet also has a lock, so that's great!

Alas, the chartfx DataSet also includes

double[] getValues(final int dimIndex);

which would force us to copy our data into an array, which sucks, exactly what we want to avoid.

The other concern is the size of the scene graph. As the scene graph grows, display updates slow down. Open a few displays with many text updates or LEDs, then call "Split Left/Right" from the tab's context menu. It can take several seconds.
So our plot aims to be just one ImageView, a single node in the scene graph, NOT one line node for each tick mark, one text node for each axis label, one circle node for each plotted point, ... And we want to prepare those plot images in background threads, then update the UI to show the most recent plot image.

The chart-fx plots seem to use several Canvas nodes. See Canvas*.java under https://github.com/ControlSystemStudio/phoebus/tree/master/app/display/representation-javafx/src/test/java/org/csstudio/display/builder/representation/javafx/sandbox for our initial tests using Canvas.
Drawing to a canvas on the UI thread means you block the UI thread. Don't want that. It looked like we can draw on the canvas off the UI thread, but turned out those drawing operations were simply buffered up and then one of two things would happen: The list of queued-up drawing operations exceeds some limit and you're dead, or they're executed on the UI thread once the canvas is added to the scene graph, so you're back to working on the UI thread.
In the end, drawing into an off-screen image and then showing that in an ImageView was the most practical option.

You'd have to check how large a scene graph is created by the chartfx plot, and if they found a way to prepare them off the UI thread.

Bottom line, by all means go ahead and test this out, and if the results are good, please update the data browser and the plot widgets to use it.

[kasemir]

Their IPAC paper states that ".. Java2D .. is at least a factor of two better than .. JavaFX (Canvas)", https://github.com/fair-acc/chart-fx/blob/main/docs/THPRB028.pdf
We obviously need to use JavaFX because that's the Phoebus graphics framework, but we do prepare the plot images with Java2D java.awt.image.BufferedImage, java.awt.Graphics2D, ..., then convert that AWT/Java2D image into a JavaFX image to show on the screen.

[flef]

Thank you for your response, I will implement it in a new branch.

See #3362.