Docs: Clarify BufferGeometry.groups.

[Mugen87]

Fixed #22555.

Description

I'd like to clarify the policy of using BufferGeometry.groups:

When groups are defined, it is invalid when parts of the vertex data are in groups but others are not. Such partial definitions are problematic for logic operating on group data. It is also not valid when groups overlap.

[donmccurdy]

Such partial definitions are problematic for logic operating on group data.

That's a surprise to me - I'd assumed groups could be resized smaller (like .drawRange) safely. Are you referring to renderer logic, exporter logic, or runtime operations like bounding boxes and raycasting?

For another example, is it OK if all indices are used but they only point to only a subset of the vertices? This may need to be explained in a bit more detail.

[Mugen87]

I've planned to add a helper for converting a multi-material mesh into several meshes. This feature requires merging of groups for a proper implementation. Not having the above policy makes an implementation noticeably more complex. Hence, I'd like to have a more strict usage of groups and avoid unnecessary complexity.

[Mugen87]

For another example, is it OK if all indices are used but they only point to only a subset of the vertices?

What is the use case of such a setup? TBH, I do not want to honor use cases which are theoretically possible just because the API currently allows it but are not relevant in production. There should always be a compelling reasons why edges cases should be supported.

[donmccurdy]

What is the use case of such a setup? TBH, I do not want to honor use cases which are theoretically possible ...

This is a very real use case, at least for geometry that does not use groups. For example, I might define a large vertex buffer, and then have several meshes drawn with different materials, each using indices that point to a subset of the vertex buffer. Is that a problem for the helper you want to add? Or only a problem if the mesh has ≥2 groups? GLTFLoader can create meshes like this, but never creates groups.

Personally I'm lukewarm on multi-material meshes and geometry.groups. I feel like they might be more maintenance burden than they're worth, and don't pair well with APIs like InstancedMesh or BatchedMesh. I certainly don't feel like grouped geometry needs to be supported in every API.

[Mugen87]

Is that a problem for the helper you want to add?

No it isn't. As long as the groups cover the entire index (so there are no gaps).

WebGPURenderer does not support groups/multi materials and there is no plan so far to add this feature. The helper should make it possible to convert multi-material meshes to a representation that can be handled by both renderers. It is also useful for exporters, see #22555.

Eventually, I hope we can stop multi-material support in WebGLRenderer at some point. The helper is then required in any case.

[donmccurdy]

I wonder if it would be worthwhile to log some kind of warning when the renderer detects a geometry that doesn't meet this requirement. That might help us to discover any use cases we've missed before enforcing this rule. I can't think of use cases myself but it would surprise me if there weren't a few...

In any case, I'm comfortable with this PR. 👍

[Mugen87]

I wonder if it would be worthwhile to log some kind of warning when the renderer detects a geometry that doesn't meet this requirement.

I'm afraid it will be a bit problematic to detect such geometries since a reliable check needs to be done per frame. And the check itself is not that trivial. Right now, I recommend to not enhance the renderer with validation checks and wait how many users actually complain about the upcoming BufferGeometryUtils.mergeGroups() and SceneUtils.createMeshesFromMultiMaterialMesh().

[donmccurdy]

Sounds good to me!

[Mugen87]

Since #23756 has been merged, this PR can be merged, too.

[iuriiiurevich]

@Mugen87
I am a bit confused... Isn't it ok to intersect geometry groups?
So this pattern is now considered bad?
Why? As far as I can see it works well.

By the way, thank you for all your work!

[Mugen87]

Yes, please don't do that anymore. The reason for this is mentioned above. I would recommend to not use groups at all but model meshes independently. You can still share geometry or buffer attribute data between them.