[Merged by Bors] - Add a scene viewer tool

[superdump]

Objective

• Allow quick and easy testing of scenes

Solution

• Add a scene-viewer tool based on load_gltf.
  • Run it with e.g. cargo run --release --example scene_viewer --features jpeg -- ../some/path/assets/models/Sponza/glTF/Sponza.gltf#Scene0
• Configure the asset path as pointing to the repo root for convenience (paths specified relative to current working directory)
• Copy over the camera controller from the shadow_biases example
• Support toggling the light animation
• Support toggling shadows
• Support adjusting the directional light shadow projection (cascaded shadow maps will remove the need for this later)

I don't want to do too much on it up-front. Rather we can add features over time as we need them.

[mockersf]

It would be nice to be able to spawn a useful default camera when there isn't one in the scene. This would be more useful than a movable camera I think

For example https://gltf-viewer.donmccurdy.com computes a bounding box, then moves the camera to be sure the box is fully visible

[mockersf]

I think this should be in /examples/tools rather than in /tools

[piaoger]

Do you plan to add environments as above gltf-viewer?

[superdump]

I think this should be in /examples/tools rather than in /tools

Done.

[superdump]

It would be nice to be able to spawn a useful default camera when there isn't one in the scene. This would be more useful than a movable camera I think

For example https://gltf-viewer.donmccurdy.com computes a bounding box, then moves the camera to be sure the box is fully visible

It depends heavily on the model. All the interesting parts of Sponza are on the inside and if you can't maneuver around to look at it then you're stuck looking at the outside walls. That said, it may absolutely make for a sensible default position in general and would allow configuring any directional light shadow projection automatically. So I'll do it. :)

[superdump]

It would be nice to be able to spawn a useful default camera when there isn't one in the scene. This would be more useful than a movable camera I think
For example https://gltf-viewer.donmccurdy.com computes a bounding box, then moves the camera to be sure the box is fully visible

It depends heavily on the model. All the interesting parts of Sponza are on the inside and if you can't maneuver around to look at it then you're stuck looking at the outside walls. That said, it may absolutely make for a sensible default position in general and would allow configuring any directional light shadow projection automatically. So I'll do it. :)

This was very awkward to do. I have to store a Handle<Scene> to wait for the scene to have loaded. Then once it has loaded, I can iterate over all the cameras in the scene world and take the first one I find that is named either of the CameraPlugin::CAMERA_2D or CameraPlugin::CAMERA_3D strings. Then spawn the scene. Then in a separate system, when I see that the scene has loaded, if there was no camera, check whether the meshes have Aabbs (they are added by the calculate_bounds system), if not, return. Otherwise, roughly transform the Aabbs to obtain a new Aabb that contains the entire scene. Use the centre and size of this to focus on the scene and have it fully in-view. Also, that the internal component and resources types are not all registered is a pain. I couldn't register my custom CameraController (from the shadow_biases example) because KeyCode is not registered.

EDIT: Updated on top of main with the camera marker changes.

There is a known issue currently where the scene does not always display. I don't know why. Sometimes it does, sometimes it doesn't. Weird.

[superdump]

Do you plan to add environments as above gltf-viewer?

Environment maps and other image-based lighting techniques will likely come in the future and then be added to this example tool, yes. I don’t know when though.

[superdump]

Well, I'm not sure why the gltf scenes sometimes load and sometimes not, but it doesn't look like it's the fault of this example tool to me. :/ bevy_gltf loads all the textures itself within its LoadContext, and I see no errors. Very confusing.

[superdump]

Help welcome for figuring out the "sometimes doesn't display the model" bug. It seems to be that the AssetServer indicates that the Handle<Scene> is loaded but the Handle<Scene> does not exist in Assets<Scene>.

[superdump]

Help welcome for figuring out the "sometimes doesn't display the model" bug. It seems to be that the AssetServer indicates that the Handle<Scene> is loaded but the Handle<Scene> does not exist in Assets<Scene>.

That wasn't it. Using commands.spawn_scene() gives really poor visibility of when the scene has actually been spawned and so when I can do things that require the scene to have been fully spawned. I had to change to using SceneSpawner::spawn(...) -> InstanceId and SceneSpawner::instance_is_ready(id: InstanceId) -> bool. Now it spawns Sponza reliably.

[mockersf]

self plug for #2424 to remove the spawn_scene command 😄

[mockersf]

Very nice to be able to quickly view any gltf file 👍

[alice-i-cecile]

Some nits, but I think this should exist. Seems quite useful, and the code itself may end up being useful to some beginners.

[alice-i-cecile]

bors r+

[bors]

Pull request successfully merged into main.

Build succeeded:

• build-android
• build (nightly, ubuntu-latest)
• build (stable, macos-latest)
• build (stable, ubuntu-latest)
• build (stable, windows-latest)
• build-wasm (nightly, ubuntu-latest)
• build-wasm (stable, ubuntu-latest)
• check-benches
• check-doc
• check-markdown-links
• check-missing-examples-in-docs
• check-unused-dependencies
• ci
• markdownlint
• run-examples