Changelog & Roadmap

A release-oriented summary of Version 2, its stability boundaries, and the remaining public-release verification steps.

Version 2.0.0

Current package line

The publishable manifest, standalone starters, changelog, and documentation now target 2.0.0. The tagged npm publish and public post-release checks remain operator steps in RELEASING.md.

Main API

• Robust finite screen joins (miter, bevel, round with fold: 'geometry' | 'exact' | false), analytic edge antialiasing (antialias( true )), explicit endpoint/closed metadata, near-plane guards, and normalized arc-length progress.
• Full shadow casting and receiving (shadow( true, true ) / setReceiveShadow( true )) across both billboards (using 3D cylindrical tube normal reconstruction and world-space shadow bias) and world-space surfaces.
• Camera-independent road/ribbon surfaces with transport or fixed-up frames, banking, distance UVs, PBR maps, true bevel/round joins and caps, and explicit reversal splits.
• GPU storage-buffer position helper (storagePositions()) for zero-CPU-upload circular ring-buffer trails and line morphing, plus lineCoord for cross-ribbon silhouette effects without UVs.
• Stable-layout update helpers (setPositionRange(), setVertexValues(), setInstanceValues()) and cached unpadded bounds (O(lines) refresh on lineWidth() / miterLimit() changes).
• Lazy optional attributes and stricter validation for point data, dimensions, widths, opacity, dashes, joins, caps, and surface construction.
• Safer default frustum behavior for CSS-pixel and shader-positioned lines.

Experimental WebGPU compute

• The root package exports MeshLineCompute, an experimental WebGPU-only façade that keeps positions, velocities, base positions, seeds, and cumulative curve distance in storage buffers.
• step( renderer, delta ) dispatches separate motion and arc-length passes. Material hooks can sample the updated distance with arcLengthNode( progress ) for distance-stable dashes, bands, and pulses.
• The Compute Curves demo exercises autonomous, follow, and field modes with mouse interaction and glowing textured ribbons.

Optional native-WebGPU entries

• makio-meshline/gpu-culling adds conservative chunk AABB compaction (with optional detached cullCamera debug inspection) for measured, very large opaque lines and surfaces. It requires native WebGPU and fails open when bounds or a render pass are ambiguous.
• makio-meshline/gpu-path is an experimental point-centric renderer for one very large static open path. It adds indirect drawing, homogeneous clipping, projected finite-leg guards, bounded center-pivot fallbacks, optional local-space simplification, .wireframe() inspection, and a complete CPU MeshLine fallback.

The regular MeshLine export from the main makio-meshline entry remains the preferred API and supports Three.js r186+ through WebGPURenderer on WebGPU and WebGL2. Read Migrating from v1 to V2 for behavior changes and application checks, V2 Benchmarks for interactive MeshLine-rendered performance diagrams, and the V2 Design Contract for the geometry guarantees.

Deliberately not promoted

• The GPU path remains experimental and falls back for multi-line/closed paths, surfaces, shadows, XR, picking, instancing, maps, dashes, and shader-hook profiles.
• Local simplification is not a screen-pixel guarantee and is disabled for advanced GPU topology.
• GPU culling remains opt-in/conservative because its measured development-machine frame-time improvement stayed below the project promotion threshold.
• MeshLineCompute remains experimental and outside the public cross-backend geometry contract. It warns and skips simulation when WebGPURenderer.compute() is unavailable; regular MeshLine remains the WebGPU/WebGL2 path.

Version 2 release verification

Documentation status:

• [x] Publish a V1-to-V2 migration guide with changed behavior, compatibility ranges, and fallback expectations.
• [x] Separate the main cross-backend API from optional native-WebGPU entry points.
• [x] Record experimental limitations and link migration, design, performance, benchmarks, and release pages from navigation.
• [x] Finalize the repository CHANGELOG.md entry for 2.0.0.
• [x] Remove prerelease status notices and align the public documentation with 2.0.0.

Package and verification status:

• [x] Set the final 2.0.0 version in the publishable package manifest.
• [x] Target ^2.0.0 from standalone starters while forcing workspace builds to exercise the current source.
• [x] Synchronize the packaged src/README.md with the public feature/status wording.
• [x] Run pnpm lint, pnpm build, pnpm docs:build, and the standalone vanilla, React Three Fiber, and Vue example builds from a clean checkout.
• [x] Install the packed artifact in isolated consumers and run runtime/declaration smoke tests at the r186.0 minimum and the pinned current Three.js version.
• [x] Run browser render coverage on both WebGPURenderer backends used by the compatibility claim.
• [x] Inspect the real npm tarball and smoke-test root, gpu-culling, and gpu-path package imports through the exports map.
• [x] Keep a non-blocking compatibility canary against the moving Three.js dev branch.
• [x] Confirm optional-entry failures remain visible and fail open on unsupported WebGL2, camera, bounds, and feature profiles.
• [ ] Publish with the intended npm tag, create the matching Git tag/GitHub release, then verify a clean registry install and documentation links.

Release history

• v2.0.0: Arc-length progress, 4-vertex billboard topology with miter/bevel/round joins and geometry/exact fold modes, analytic edge antialiasing, shadow casting and receiving with 3D cylindrical tube normals, world-space road/ribbon surfaces, storagePositions() GPU trails/morphing, partial position updates (setPositionRange()), O(lines) bounds refresh, complete framework starters, optional native-WebGPU chunk culling, an experimental point-centric GPU path with CPU fallback, and experimental WebGPU compute curves.
• v1.4.0: New MeshLinePickerHelper (Three.js-style helper that overlays the GPU picker's hit-zones with a distinct hue per instance), new picker.lines snapshot getter, aligned point-array input handling, TypeScript declarations, and material shadow-node docs.
• v1.3.2: Normalize lineWidth semantics, stabilize GPU-positioned endpoint tangents, fix pre-build .opacity(), and expand options-object configuration.
• v1.3.1: Corner rendering overhaul — uniform thickness at every bend, stable miter at oblique camera angles, and CPU-side smoothing (smoothSharpBends) for very sharp polylines. useMiterLimit option removed; miter is always on.
• v1.3.0: R3F & Vue StackBlitz examples with declarative <MeshLine> wrapper components
• v1.2.3: Fixes for alphaMap(), material.copy(), and sineWavePositions(); minor allocation cleanups
• v1.2.2: Buffer usage fixes, frustum-culling skip, miter handling for near-180° angles
• v1.2.0: Shadow casting (.shadow(), castShadowNode); Shadow & Bamboo Grove demos
• v1.1.0: Three.js r181 support, faster geometry, ensureBuilt()
• v1.0.5: Fix TSL assign operation error
• v1.0.4: Per-vertex RGB colors; Three.js ^0.180.0 peer
• v1.0.3: counters→progress rename, alphaTest/transparency fix
• v1.0.0: Initial release — TSL, batching, instancing, dashes, gradients, miter, hooks, GPU positions

Full changelog: CHANGELOG.md

Roadmap

• Mid-term:
  • Extend MeshLineCompute with real texture/audio/depth force integrations and compute-aware segment picking.
  • Promote the optional GPU path's variable topology after multi-line/closed, shadows, XR, and picking gates.
• Long-term:
  • Explore optional compute pre-passes for dynamic polylines only where resampling, tangent generation, or arc-length work produces a measured end-to-end win.
  • Extend the experimental point-centric GPU path with integrated segment-run culling, screen-error LOD, closed/multi-line topology, shadows/XR, and transported road frames.
  • Add advanced PBR surface customization and instancing without coupling it to billboard-only TSL hooks.
  • Explore further WebGPU performance opportunities only where matched benchmarks show a material win.