GltfLoader issues

[capdevon]

The recent changes to the GLTF loader have introduced several problems:

1. Duplicate Controllers: The new loader creates duplicate AnimComposer and SkinningControl controllers for each geometry in the model. (see here)
2. Nested Geometry: Each geometry is now encapsulated within two parent nodes, which can complicate scene hierarchy management.
3. Backward Compatibility: These changes may cause compatibility issues with existing code that relies on the previous behavior of the GLTF loader. (see here)

Other issues reported by adi.barda: (see here)

• Blured colors - maybe normal map is missing?
• Sometime missing animations - maybe related to 1

[stephengold]

Please provide detailed instructions for reproducing this issue.
Has anyone identified which recent changes caused this issue?

[capdevon]

Maybe this: #2103

[stephengold]

Would reverting 2103 solve this issue?

[capdevon]

Would reverting 2103 solve this issue?

Probably yes.

Please provide detailed instructions for reproducing this issue.

Here is the 3D Model for the test file

jME-3.7.0

jME-3.6.1-stable

[stephengold]

Using a simple test app, I showed that the issue of YBot loading with multiple AnimComposers pre-dates PR 2103. I'll do a bisection search to find where the issue was introduced.

Here is the test app:

import com.jme3.anim.AnimComposer;
import com.jme3.app.SimpleApplication;
import com.jme3.scene.Node;
import com.jme3.scene.Spatial;
import com.jme3.scene.control.Control;
import com.jme3.scene.plugins.gltf.GltfModelKey;
import java.util.List;

public class TestYBot extends SimpleApplication {

    public static void main(String[] args) {
        TestYBot app = new TestYBot();
        app.start();
    }

    @Override
    public void simpleInitApp() {
        GltfModelKey modelKey = new GltfModelKey("YBot.gltf");
        Spatial modelRoot = assetManager.loadModel(modelKey);
        int numComposers = countControls(modelRoot, AnimComposer.class);
        System.out.println("numComposers = " + numComposers);
        stop();
    }

    /**
     * Count all controls of the specified type in the specified subtree of a
     * scene graph. Note: recursive!
     *
     * @param <T> subclass of Control
     * @param subtree the subtree to analyze (may be null, unaffected)
     * @param controlType the subclass of Control to search for
     * @return the count (&ge;0)
     */
    private static <T extends Control> int countControls(
            Spatial subtree, Class<T> controlType) {
        int result = 0;

        if (subtree != null) {
            int numControls = subtree.getNumControls();
            for (int controlI = 0; controlI < numControls; ++controlI) {
                Control control = subtree.getControl(controlI);
                if (controlType.isAssignableFrom(control.getClass())) {
                    ++result;
                }
            }
        }

        if (subtree instanceof Node) {
            Node node = (Node) subtree;
            List<Spatial> children = node.getChildren();
            for (Spatial child : children) {
                result += countControls(child, controlType);
            }
        }

        return result;
    }
}

[stephengold]

The issue was introduced by PR #2060.

With "master" branch at f915e56, the test app prints numComposers = 1.
With "master" branch at 4c87531, the test app prints numComposers = 2.

[riccardobl]

2. Nested Geometry: Each geometry is now encapsulated within two parent nodes, which can complicate scene hierarchy management.
3. Backward Compatibility: These changes may cause compatibility issues with existing code that relies on the previous behavior of the GLTF loader. (see here)

As pointed out in the pr #2103 , this is to solve several issue and inconsistencies with the scenegraph of imported models , it is not backward compatible, but the previous behavior was bugged and source of issues

[stephengold]

The current behavior, with multiple anim composers per model, is clearly bugged. And that change appears to be unrelated to PR 2103.

[stephengold]

I think there's just one unsolved piece to this issue: the "blurred colors" change, which occurred between v3.6.1-stable and v3.7.0-beta1.2.2.
@scenemax3d sent me a copy of his test app, which I've now run with both releases. I hope to use a bisection search to identify the commit(s) that caused the change.

[stephengold]

I simplified the test app so it doesn't require Minie:

package jme3test;

import com.jme3.app.SimpleApplication;
import com.jme3.light.AmbientLight;
import com.jme3.light.DirectionalLight;
import com.jme3.math.ColorRGBA;
import com.jme3.math.Quaternion;
import com.jme3.math.Vector3f;
import com.jme3.scene.Node;
import com.jme3.scene.Spatial;
import com.jme3.shadow.DirectionalLightShadowRenderer;
import com.jme3.system.AppSettings;

public class TestApp extends SimpleApplication {

    public static void main(String[] args) {
        TestApp app = new TestApp();
        app.showSettings = false;
        AppSettings settings = new AppSettings(true);

        settings.setWidth(1024);
        settings.setHeight(768);
        settings.setGammaCorrection(false);
        settings.setSamples(4); // anti-aliasing
        settings.setVSync(true);
        app.setSettings(settings);
        app.start(); // start the game
    }

    @Override
    public void simpleInitApp() {
        flyCam.setDragToRotate(true);
        cam.setLocation(new Vector3f(5.809729f, 3.3733122f, 4.834249f));
        cam.setRotation(new Quaternion(0.0018560609f, 0.9064233f, 0.0039831875f, -0.42234766f));

        loadModel1();
        addLighting();
    }

    private void addLighting() {
        DirectionalLight sun = new DirectionalLight();
        sun.setColor(ColorRGBA.White);
        sun.setDirection(new Vector3f(-0.1f, -0.7f, -1.0f));
        rootNode.addLight(sun);

        final int SHADOWMAP_SIZE = 1024;
        DirectionalLightShadowRenderer dlsr = new DirectionalLightShadowRenderer(assetManager, SHADOWMAP_SIZE, 3);
        dlsr.setLight(sun);
        viewPort.addProcessor(dlsr);

        AmbientLight al = new AmbientLight();
        al.setColor(ColorRGBA.White.mult(0.3f));
        al.setEnabled(true);
        rootNode.addLight(al);
    }

    private void loadModel1() {
        Spatial mm = assetManager.loadModel("Models/boss/boss.gltf");
        final Node model = new Node();
        model.attachChild(mm); // add it to the wrapper
        mm.scale(3f, 3f, 3f);
        rootNode.attachChild(model);
    }
}

[stephengold]

The "blurred colors change" had already occurred as of efbcd03 (29 May 2023 in the "master" branch), which means it wasn't related to PR #2060.

It hadn't occurred yet as of fec08b3 (11 Feb 2023 in the "master" branch). So now I've got a 15-week window to search...