@mxpv — I'd like to tackle value clips next. They're the big remaining gap in §12.3 value resolution, and about the last major piece before the crate can resolve time-varying data the way most production assets actually ship it (clips are how USD streams crowds/sims/caches).
The AOUSD supplemental release vendored under vendor/ includes a reference value-resolution implementation plus conformance assets (clip_basic / advanced / multi / sets / timings), so I'd validate directly against those.
High-level plan:
- A core PR for explicit clips: the
clips / clipSets metadata, stage↔clip time mapping, active-clip selection, manifest gating, and slotting clips into resolution at the spec-correct strength (weaker than local layer-stack opinions, stronger than opinions across references; clips supply only time-varying values, never defaults/metadata). Passes the full vendored conformance suite and flips the ROADMAP row.
- A follow-up PR for the parity pieces the reference doesn't cover: template clips,
interpolateMissingClipValues, and UsdClipsAPI authoring.
The one part that reaches into the composition core is value resolution itself: clips resolve as a per-node value source, so timeSamples resolution has to walk the PCP nodes rather than use the current flattened query. Wanted to flag that and check the direction before I start.
Does this fit where you want the crate to go, and does the approach sound right?
@mxpv — I'd like to tackle value clips next. They're the big remaining gap in §12.3 value resolution, and about the last major piece before the crate can resolve time-varying data the way most production assets actually ship it (clips are how USD streams crowds/sims/caches).
The AOUSD supplemental release vendored under
vendor/includes a reference value-resolution implementation plus conformance assets (clip_basic / advanced / multi / sets / timings), so I'd validate directly against those.High-level plan:
clips/clipSetsmetadata, stage↔clip time mapping, active-clip selection, manifest gating, and slotting clips into resolution at the spec-correct strength (weaker than local layer-stack opinions, stronger than opinions across references; clips supply only time-varying values, never defaults/metadata). Passes the full vendored conformance suite and flips the ROADMAP row.interpolateMissingClipValues, andUsdClipsAPIauthoring.The one part that reaches into the composition core is value resolution itself: clips resolve as a per-node value source, so timeSamples resolution has to walk the PCP nodes rather than use the current flattened query. Wanted to flag that and check the direction before I start.
Does this fit where you want the crate to go, and does the approach sound right?