RFD 1061: GLB upload prep moves to idtx_core, later
Problem
__tests__/prepareGlbForApiUpload.test.js tested two functions, computeApiUploadSimplifyRatio and prepareGlbForApiUpload, that glbCompress.js never defined. TaskManager.jsx’s skintokens auto-rig path imports prepareGlbForApiUpload at line 1163 and awaits it, and the import threw since whenever that call site was written. RFD 1023 first recorded this gap, and it stayed open through RFD 1060’s move.
The mesh a browser uploads today is prepared by compressGlbBuffer in the same file, over gltf-transform: client-side Draco, Meshopt, and WebP. RFD 1053 makes OpenUSD the internal format, the way .blend is internal to Blender, and it names glTF only a transmission format, converted at the boundary. GLB, Draco, and WebP belong at that boundary. They do not belong as the mesh-prep pipeline itself, which is what compressGlbBuffer is today.
Decision
The upload-prep job belongs on idtx_core’s USD-native transport, not on a second bespoke GLB pipeline in JavaScript. thirdparty/fabric-flow-adapters/flow/ already carries what that job needs, and this repository’s browser compression does not.
None of that is reachable from a browser today. See DETAILS.md for what idtx_core carries, the three steps to reach it, the interim stopgap in glbCompress.js, and a bug that stopgap’s tests surfaced.