@displayxr/inline3d 1.42.0 → 1.43.0

package/CHANGELOG.md

@@ -5,6 +5,55 @@ entry points (`.`, `./three`) are frozen for 1.x, while the **scene subpaths** (
 `./splat`, `./model`) are a preview tier whose options may change in any release. Entries below say
 which tier they touch, because that is what tells you whether an upgrade can move your pixels.
 
+## 1.43.0 — 2026-10-12
+
+### `./splat` (PlayCanvas) — `warmEffect`: no frozen frame when a reveal first plays
+
+- **`handle.warmEffect(name, opts)`** compiles the program `playEffect(name, opts)` will need and
+  links it in idle time (`KHR_parallel_shader_compile`), so the effect's first frame no longer
+  blocks on the link. On a Windows panel `'assemble'` froze for 250–300 ms on every cold session.
+  Resolves `true` once warm, `false` when nothing could be warmed (entity scope, custom fragment
+  stage). It is also on the deferred handle, so it can be called right after `addSplat`.
+- **`prepareSource(src, { warmEffects: [...] })`** warms effects you will play once that asset is
+  in (names or `{ name, opts }`), after its load and its transition pre-warm.
+- docs/splat-effects.md §Warming an effect.
+
+### `./splat` + `./model` (PlayCanvas) — `warmEntity`: a glTF's first draw without the freeze
+
+- **`handle.warmEntity(entity)`** (splat handle: content under `handle.engine.root`; model
+  handle: defaults to the model) compiles the entity's lit programs in idle time through the
+  renderer's own `MeshInstance.getShaderInstance`, with its exact pass, light hash, lights, view
+  format and the eye cameras' live shader params. It then links them off the frame
+  (`KHR_parallel_shader_compile`, time-sliced issuing), so the first draw no longer blocks.
+  PlayCanvas links every new lit program synchronously on first draw: a 13-material car froze
+  0.8–0.95 s on a panel. On Khronos CarConcept (13 programs), headless: 0 links on show (from 13),
+  worst frame 20 ms (from 102 ms with the shader cache warm, 2.69 s cold), pixel-identical. Spark
+  and three resolve 0. docs/playcanvas-adapter.md §warmEntity.
+
+## 1.42.1 — 2026-10-12
+
+Touches the **preview tier** (`./call`, `./camera`) only: auto-convergence and the stereo-tile depth
+cursor are steadier on a wide-baseline camera. No API change.
+
+### Auto-convergence holds the face on a 120 mm-baseline camera
+
+- Found on the SR laptop's built-in stereo camera (120 mm baseline, grayscale): auto-convergence
+  flipped between the face and the background. On a dim frame the sparse block grid (1.5 blocks
+  apart) matched ~9 blocks, 3-4 on the face, too few to form the subject mode, so the background
+  won. The grid is now one block apart (~36 blocks: the face on every frame, ±1 px instead of ±4),
+  and a mode's bin widens with its disparity (10%), so a near face's own relief stays one group.
+- The convergence shift may now reach 20% of the eye width (was 12%): that camera sees a face at
+  0.6 m at ~22% of the eye width, ~11% per eye, which left no room to lean in.
+
+### Stereo-tile depth cursor: steady at foreground edges
+
+- The footprint was matched from the left eye's image only. At a foreground edge the left eye's
+  block can sit on the subject while the right eye's sits on the background, and the arrow jumped
+  out and back frame to frame. Each footprint point is now matched from BOTH eyes, every match must
+  pass a left-right consistency check (matched back, it lands within 1 px of where it started), and
+  the nearest value counts only when a second sample agrees; otherwise the arrow stays on the
+  glass. Over 20 live frames across a face edge: 200 → 35 large jumps.
+
 ## 1.42.0 — 2026-10-11
 
 Touches the **core** `./three` (`DepthCursor`) and the **preview tier** (`./model`, `./splat`,