Unified Mesh Loading API

[fm3]

There are currently three different APIs for meshes: precomputed v3, precomputed (legacy), ad-hoc.

Can we provide a unified API for all three?

Things to consider

• LODs
• mags
• multiple meshfiles can exist for the same segmentation
• encodings/compression (DRACO,stl,…)
• mappings (some meshfiles are computed for oversegmentation. need to request all the meshes mapped to target id)

Rough Design:

• LIST SOURCES: List available mesh sources (meshfiles, adhoc, remote?)
• METADATA + PARAMETER RANGES: Second route (or as part of first one) to query metadata for a mesh source:
  • available parameter ranges
    • Numeric (LODs, mags, subsamplingStrides)
    • Applyable Mappings (mapping name in case of mapping-specific meshfile, ANY in case of oversegmentation meshfile or adhoc mapping)
  • encoding info (DRACO,stl,…)
  • needs NEIGHBORS logic
• LIST CHUNKS
  • For a mesh source, and a segment/agglomerate id, a mag, an LOD, a mapping name, list available chunks (returns transform, offset and for each chunk a position)
• CHUNK DATA
  • data for a list of chunks (as identified by mesh source, segmentId/agglomerateId, mappingName, lod, mag, position)
  • may return more chunks than are requested (in case of mapping and oversegmentation meshfile)
  • needs to return metadata as well (byte lengths of returned chunks + counts per position)

Problem:

• NEIGHBORS case
  • no list chunks, instead data by single seed position (from click), response has NEIGHBORS header

Question:

• request multiple LODs in one request?
• tracingstore vs datastore? should both offer this route? frontend needs to distinguish? (meshfiles are only on datastore. for editable mappings, datastore requests segment list from tracingstore)

Alternative thought

• new route for loading the whole mesh (backend puzzles it together)
• or just use the old routes in libs

[fm3]

Summary of decisions from meeting:

• add routes for loading a whole mesh (backend puzzles it together, serves stl)

• drop support for “old” meshfile format, perform migration for existing files

• frontend can continue to use existing chunked data loading routes

• python client should download whole mesh, using seed positions passed from frontend in case of ad-hoc without segment index