WebGPU 3D tier + full backend parity, spine on WebGPU — 20.0.0 (#1184, #1536)

[obiot]

Completes the WebGPU renderer (#1184): the full 3D tier, the remaining backend-parity surface, point/spot 3D lights (#1536), custom mesh shaders on both GPU backends, and a WebGPU spine batcher with full two-color tinting. With this, video.AUTO defaults to WebGPU → WebGL 2 → Canvas and every engine feature renders identically on both GPU backends (all 44 examples verified side-by-side, headed).

Engine (packages/melonjs)

3D tier on WebGPU

• drawMesh through unlit/lit WGSL mesh pipelines: retained model-space geometry under Camera3d (upload once; placement/tint/alpha-cutoff/emissive ride one per-draw uniform snapshot) plus the CPU-projected 2D-camera path; per-mesh cull/winding as pipeline state; supportsDepthBuffer / supportsRetainedMesh now true
• depth realized as the render pass's load/store ops — one depth clear per target per frame (GL policy); pure-2D applications keep byte-identical passes
• glTF scenes/animation, Sprite3d billboards and split-screen Camera3d viewports run unchanged on top

Lighting (#1536)

• Light3d gains "point" and "spot" on both backends: 12-float std140 light entries, quadratic-over-range falloff (the Light2d model — deliberately not physical inverse-square), smooth-edged cones, runtime-mutable fields
• glTF loader instantiates all three punctual types (scene-scaled position/range/cones), copies authored light names onto the Light3d (getChildByName("Sun")), and gains a lightIntensityScale load option; lamp-only scenes now correctly flag their meshes lit

Custom mesh shaders

• GLShader is now dual-language: new GLShader(renderer.gl, { vertex, fragment, wgsl }) with isWebGL/isWebGPU flags; mesh.shader hosts it on either backend, warn-and-degrade on mismatch (WGSL validation is async and contained — a bad module can never poison frame submission)
• shader assets grow the matching complete-program shape { type: "shader", src: { vertex, fragment, wgsl } }

Completion + parity batch

• antiAlias: true → 4× MSAA on WebGPU canvas passes; mesh textures gain generated mip chains + trilinear + 4× anisotropy on both backends (authored compressed chains included; "nearest" opts out); 8-slot multi-texture quad batching
• WebGL gains setBlendMode("none") and the frameless-video upload guard; WebGPU gains enableScissor/setBlendEnabled/clearRenderTarget, settings.batcher, startup settings.blendMode, GPUVendor, failIfMajorPerformanceCaveat (rejects software adapters)
• ShineEffect adopts the uUVYDir seam (vertical sweeps ran backwards on the GL pooled path — same family as the DropShadow fix)
• shared neutral hoists (gpu/meshvertex.ts, gpu/quadcorners.ts, gpu/primitives.ts) delete ~250 duplicated lines across backends
• published typings hygiene: renderer internals are @internal-tagged and stripped from the emitted .d.ts by a new build pass

Docs

• JSDoc + README refresh across the 3D surface (retained-mesh story, capability flags, dual-language shaders, glTF light options, AUTO ladder); the wiki was updated in lockstep

Spine plugin (packages/spine-plugin, 4.0.0)

• new WebGPUSpineBatcher: spine-core's backend-neutral geometry pass feeds a WGSL port of the official two-colored shader — dark-tint now works on WebGPU, and mesh-heavy skeletons batch instead of drawing per-triangle
• backend dispatch via renderer.type; SpineBatcher → WebGLSpineBatcher (internal rename); spine runtimes ^4.3.7 → ^4.3.13

Examples

• all examples swept to video.AUTO; the 3D examples guard on supportsDepthBuffer; reworked Hello WebGPU

Verification

• suite 5528 passed / 9 skipped (≈370 new tests across the arc), eslint/biome/tsc clean
• three review rounds (3D-tier code review, shared-architecture review, documentation review) — all findings fixed; the one deliberate deferral is instancing (3D: mesh instancing (drawElementsInstanced for repeated geometry) #1508), an API-design ticket
• headed side-by-side WebGL↔WebGPU: all 44 examples, the 9 3D examples in depth, spine's 18 skeletons on all three backends

🤖 Generated with Claude Code

https://claude.ai/code/session_01QVjYzf76AEU3wJk766JAQi

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.