Handle multiple feature ID sets in ModelExperimental

[ptrgags]

Fixes #10007
Fixes #9894

This also partially addresses #9884, though picking is not currently supported for feature IDs without a property table, and the current picking system would need larger changes to address this (see #9884 (comment))

Summary of changes:

• In ModelComponents, combined featureIdAttributes and featureIdTextures into one featureIds array. This is more consistent with the current EXT_mesh_features spec.
• in ModelExperimental, replaced featureIdAttributeIndex and featureIdTextureIndex with featureIdIndex and instanceFeatureIdIndex, as the latter pair better match the glTF structure
• FeatureIdPipelineStage now processes all the feature ID attributes instead of only the selected one. It populates a FeatureIds struct that can be passed into both feature ID selection (next bullet) and custom shaders.
• SelectedFeatureIdStage handles the shader details for taking the selected feature ID set and populating a struct used by styling and picking. This is significantly easier now with the FeatureIds struct, it's simply selectedFeature.id = featureIds.SELECTED_FEATURE_ID (using a define to substitute in the right variable).
• adds feature IDs to Custom Shaders via fsInput.featureIds.featureId_N or fsInput.featureIds.instanceFeatureId_N
• updated the 3D Tiles Next Photogrammetry Sandcastle to use the new fsInput.featureIds.featureId_N syntax to avoid exposing private API.

To Do:

• Update custom shader guide
• Update documentation for feature ID stages
• Fix broken unit tests
• Add new unit tests
• add featureIdIndex and instanceFeatureIdIndex to Cesium3DTileset for easier configuration
• Update CHANGES.md
• Check if instanced FEATURE_ID_n is added to the ProcessedAttributes correctly
• fix race condition about featureIdIndex being set before the scene graph is constructed.
• investigate styling, sometimes higher LODs aren't displaying the right color
• Update other existing sandcastles
• Investigate ModelColorPipelineStage DeveloperError
• Investigate draco tile failures in Custom Shaders 3D Tiles sandcastle -> opened Fix regression in GltfLoader for handling Draco _BATCHID #10031

[cesium-concierge]

Thanks for the pull request @ptrgags!

• ✔️ Signed CLA found.
• ❔ CHANGES.md was not updated.
  • If this change updates the public API in any way, please add a bullet point to CHANGES.md.

Reviewers, don't forget to make sure that:

• Cesium Viewer works.
• Works in 2D/CV.

[javagl]

Just a short note: I think that http://localhost:8080/Apps/Sandcastle/index.html?src=3D%20Tiles%20Next%20S2%20Globe.html&label=3D%20Tiles%20Next is still using the (now no longer working, internal) FEATURE_ID_TEXTURE approach for the custom shader.

[ptrgags]

@lilleyse this PR is now ready for review. I do still want to add a sandcastle to show multiple feature Id sets once you finish updating the Ferry Building tileset and upload it to ion

[ptrgags]

@lilleyse I updated the 3D Tiles Next Photogrammetry Sandcastle with new styling and custom shader examples

[lilleyse]

Seems to be in pretty good shape overall. Here's my first round of feedback, I plan to test some examples next.

[lilleyse]

@javagl just to confirm, have you tested all the samples in CesiumGS/3d-tiles-samples#40 on this branch?

Here's the testing strategy I was going to go with. Are there any that can already be checked off?

• Test picking feature without corresponding property table (expect it not to pick anything until #9884 is addressed)
• Test setting the active feature id and then picking and styling
• Test setting the active instance feature id and then picking and styling
• Test picking pnts with BATCH_ID
• Test picking pnts with per point properties (expect picking to work now, but maybe not in the future #10025)
• Test picking i3dm
• Test picking b3dm
• Test picking glTF with feature ID attributes
• Test picking glTF with feature ID textures
• Test picking glTF with implicit feature IDs
• Test picking glTF with instance feature ids
• Test picking glTF with implicit instance feature ids
• Test custom shader for any of the above

[ptrgags]

Addressed the concrete feedback. Next step is investigating a bug @lilleyse pointed out to me where switching between a style and a custom shader leaves the tileset in a bad state:

[ptrgags]

@lilleyse this is ready for review again.

[ptrgags]

merging in the let/const changes didn't seem to work properly... I'm going to reset my branch and try again

[lilleyse]

The new sandcastle option Multiple Feature ID Sets might be a little too busy with the color scheme. Maybe it's better to go with something simpler, like tinting roof windows yellow and facade windows purple.

[lilleyse]

Aside from that, the recent changes look good. @javagl once you're satisfied with your testing we can get this merged.

[ptrgags]

@lilleyse I simplified the sandcastle like you suggested

[javagl]

I did some tests, roughly based on the bullet points above, but with some disclaimers (summarized below)

• Test picking feature without corresponding property table (expect it not to pick anything until #9884 is addressed)

  • This seems to work. based on the ExplicitFeatureIds and ImplicitFeatureIds samples from Added basic EXT_mesh_features samples 3d-tiles-samples#40, which both define feature IDs without associated properties.

• Test setting the active feature id and then picking and styling

  • This seems to work, based on the (preliminary) ComplexExample mentioned in Added basic EXT_mesh_features samples 3d-tiles-samples#40 (comment)

• Test setting the active instance feature id and then picking and styling

• Test picking pnts with BATCH_ID

• Test picking pnts with per point properties (expect picking to work now, but maybe not in the future #10025)

  • I think these are working as expected. When loading the cesium/Apps/SampleData/Cesium3DTiles/PointCloud/PointCloudBatched in my test sandcastle, it shows the "feature ID" (aka Batch ID) and some per-point properties in the Tooltip, if one manages to move the mouse directly over that single pixel :D

• Test picking i3dm

• Test picking b3dm

  • These also seem to work, based on the Instanced/InstancedWithBatchTable and Batched/BatchedWithBatchTable sample data...

• Test picking glTF with feature ID attributes

• Test picking glTF with feature ID textures

• Test picking glTF with implicit feature IDs

  • These seem to work based on the ExplicitFeatureIdsAndSimpleProperty, ImplicitFeatureIdsAndSimpleProperty and SimpleFeatureIdTexture samples from Added basic EXT_mesh_features samples 3d-tiles-samples#40

• Test picking glTF with instance feature ids

• Test picking glTF with implicit instance feature ids

• Test custom shader for any of the above

---

Disclaimers:

• In general, I wasn't sure about the interplay of metadata and styling. For example, there is this line with color: "color(${color})" in one of the samples, and I don't know what color is. (No not that one, the other one. I mean the innermost one...)

• Setting a style with a pointSize: 10.0 does not seem to have any effect:

  tileset.style = new Cesium.Cesium3DTileStyle({
    pointSize: 10.0
  });
  

  Is this a known issue?

• Everything related to instance IDs: I'll have to look into that, in terms of example data and usage.

• Custom shaders: Also would need a bit more time here to try out "things"...

---

General:

I've been testing most of this based on the (sketchy) "Test Sandcastle" that may, at some point, be a "template" for the example sandcastles that will go into the READMEs for the samples.

There are several issues with that:

Right now, accessing the metadata generically is ... clumsy: It has to do many instanceof/undefined checks, and a lot of "implicit, specific knowledge" is carved into this code. It might be that there are some abstractions in the latest state of this PR. If some of that code could be simplified, that would be great.

There is an experimental function for "Setting the active feature ID", but ... I have no idea how to know the actual number of available feature IDs in the first place. More generally: I assume that the concept of having an "active feature ID" is a tribute to having to transport this information through the rendering process into "the picking world". I could question the now 54 (fifty-four!) properties in the options for a tileset, and ask who is going to create a "mental model" of what they mean and what values they should have. But even if it is OK to just add more properties there, this is part of a "global state" (global in terms of the tileset), and I see this as a (possibly severe) limitation. What if someone wants to do styling based on one feature ID, and show a tooltip based on another feature ID? (Related to that: The ambiguities between "feature ID" and "instance feature ID" are ... not so nice. Conveying the difference and their meaning in context, in some sort of documentation/tutorial, may be difficult...)

It might be that these issues are already resolved by more recent changes. So to put it that way: If somebody could improve (as in: simplify and generalize) this test code at https://github.com/javagl/3d-tiles-samples/blob/add-ext-mesh-features-samples/EXT_mesh_features/CommonSandcastleCode.txt , that would be great...

[ptrgags]

@javagl a couple things to point out:

• Styling with property names should work for meshes, but not for point clouds, that falls under Model: GPU Styling #9944
• As far as picking goes, indeed the metadata is cumbersome to access, but this PR is only concerned with
• pointSize in Cesium3DTileStyle... 😬 I don't think that was implemented. I opened Cesium3DTileStyle.pointSize in ModelExperimental #10046

[lilleyse]

I finished up testing. All looks good.

• Test setting the active instance feature id and then picking and styling
• Test custom shader for any of the above
• Test picking glTF with instance feature ids (Sandcastle)
• Test picking glTF with implicit instance feature ids (Same sandcastle as above)