How to get an ObservableFloatArray from a Stream?

Question

This is nearly the same question as this one, but in the opposite direction.

I know there\'s no FloatStream in Java 8, and there\'re no many use cases f

Answer 1

I don't think there's any way around the fact that you have to create one data structure (e.g. a double[] or a List<Float>) and then map it into a float[]. (But maybe I am missing something.)

If you want to do this using a Stream-like API, you can use a Collector to do the mapping at the end:

import java.util.List;
import java.util.ListIterator;
import java.util.Random;
import java.util.stream.Collector;
import java.util.stream.Collectors;
import java.util.stream.IntStream;
import java.util.stream.Stream;

import javafx.scene.shape.TriangleMesh;

public class StreamFloatsTest {

    public static void main(String[] args) {

        // The following declaration works in Eclipse 4.4
        // however it won't compile from the command line:
        // Collector<Float, List<Float>, float[]> toFloatArray =

        // This declaration works:

        Collector<Float, ?, float[]> toFloatArray =
                Collectors.collectingAndThen(Collectors.toList(), floatList -> {
                    float[] array = new float[floatList.size()];
                    for (ListIterator<Float> iterator = floatList.listIterator(); iterator.hasNext();) {
                        array[iterator.nextIndex()] = iterator.next();
                    }
                    return array ;
                });


        // Random vertex list for demo purposes:
        Random rng = new Random();
        List<Vertex> vertices = IntStream.range(0, 100)
                .mapToObj(i -> new Vertex(rng.nextFloat(), rng.nextFloat(), rng.nextFloat()))
                .collect(Collectors.toList());

        float[] vertexArray = vertices.stream()
                .flatMap(v -> Stream.of(v.getX(), v.getY(), v.getZ()))
                .collect(toFloatArray);


        TriangleMesh mesh = new TriangleMesh();

        mesh.getPoints().addListener(obs -> System.out.println("mesh invalidated"));
        mesh.getPoints().addListener((array, sizeChanged, from, to) -> System.out.println("mesh changed"));
        mesh.getPoints().addAll(vertexArray);

    }

    public static class Vertex {
        private final float x ;
        private final float y ;
        private final float z ;
        public Vertex(float x, float y, float z) {
            this.x = x ;
            this.y = y ;
            this.z = z ;
        }
        public float getX() {
            return x;
        }
        public float getY() {
            return y;
        }
        public float getZ() {
            return z;
        }
        public float[] getCoordinates() {
            return new float[] {x, y, z};
        }
    }
}

Answer 2

Keep in mind that a Stream defines an operation rather that a storage. So for most operations, using a float provides only little benefit over double values when CPU registers are used. There might be a theoretical improvement for operations that could be accelerated using SSE or GPU, but that’s not relevant here.

So you can use a DoubleStream for that operation, the only thing you need is a collector capable of collecting a DoubleStream into a float[] array:

float[] arrayVertices = listVertices.stream()
    .flatMapToDouble(vertex->DoubleStream.of(vertex.x, vertex.y, vertex.z))
    .collect(FaCollector::new, FaCollector::add, FaCollector::join)
    .toArray();

static class FaCollector {
    float[] curr=new float[64];
    int size;
    void add(double d) {
        if(curr.length==size) curr=Arrays.copyOf(curr, size*2);
        curr[size++]=(float)d;
    }
    void join(FaCollector other) {
        if(size+other.size > curr.length)
            curr=Arrays.copyOf(curr, size+other.size);
        System.arraycopy(other.curr, 0, curr, size, other.size);
        size+=other.size;
    }
    float[] toArray() {
        if(size!=curr.length) curr=Arrays.copyOf(curr, size);
        return curr;
    }
}

This supports parallel processing, however, for an operation that merely consist of data copying only, there is no benefit from parallel processing.