Draco encoder, "Error compressing attribute"

I ran into this peculiar error:

Error: Error compressing "COLOR_0" attribute.

It comes from here: https://github.com/donmccurdy/glTF-Transform/blob/1ef6be94787c4c6dcb21d011d219de73b0ff0e8a/packages/extensions/src/khr-draco-mesh-compression/encoder.ts#L80

I did a bit of digging and can verify a few things:

• The attribute seems to be created on the Mesh object created by the encoder module via const mesh = new encoderModule.Mesh();, as the value of mesh.num_attributes() is 7, and this happens to be the 7th, and last one.
• attribute.getComponentType() is ‌5126 (that’s the same value as Accessor.ComponentType.FLOAT)
• attribute.getCount() is 450
• attribute.getElementSize() is 4
• attribute.getArray() is a Float32Array of size 1800, this checks out as 450 * 4

Since the result value is just -1, I’m having a bit of a difficulty understanding the reason for this failure.

[donmccurdy]

Sorry, I haven’t been able to find a way to reproduce this issue. If this error comes up again with a model you are able to share, please let me know!

[donmccurdy]

Which steps lead to a successfully converted and compressed model?

Another idea to try would be higher values (16?) of the --quantizeColor option.

I’ve tried running a batch conversion over ~50 samples I had on hand, and couldn’t reproduce this error for any of them, so I think I may need to leave this open until someone can share a reproducible case. It may be that I’ve implemented something incorrectly here, but glTF-Pipeline also pre-processes the data a bit more, so I wonder if that is just smoothing out some edge case that the Draco encoder doesn’t like.