WebGL 2 only renderer + Vertex Array Objects (#1509) — 20.0.0

[obiot]

Phase 0 of the WebGPU groundwork (#1184): drop the WebGL 1 path, and give every batcher an immutable vertex state so the layout stops being re-issued per draw and starts looking like a pipeline descriptor.

Breaking: the WebGL renderer requires WebGL 2

• renderer: video.AUTO falls back to Canvas on WebGL-1-only devices; renderer: video.WEBGL throws there
• preferWebGL1 setting and the #webgl1 URL flag are removed (#webgl / #webgl2 are synonyms)
• device.isWebGLSupported() now probes for a WebGL 2 context. This fixes a latent inconsistency: it probed WebGL 1 while context creation requested WebGL 2, so the support gate and the actual renderer could disagree.
• renderer.type is always "WebGL2"; renderer.WebGLVersion is deprecated (always 2)
• Behaviour corrections on hardware that previously fell back to WebGL 1: repeat wrap genuinely tiles NPOT textures (was clamp + warning), "darken"/"lighten" use true MIN/MAX equations (were silently downgraded to "normal"), createPattern() accepts NPOT sources (threw before)

User shaders need no changes. ShaderEffect bodies and raw GLShader sources stay GLSL ES 1.00 — WebGL 2 compiles them by spec, and every modern install already ran them on a WebGL 2 context. Engine shaders are deliberately not migrated to ES 3.00 here; the rule going forward is that a shader migrates only when it needs a 3.00 feature (first such case ticketed as #1552).

Vertex Array Objects

• New WebGLVertexState (buffer/vertexstate.js, sibling of WebGLIndexBuffer) owns the VAO lifecycle. Its binding save/restore is private, so no caller can half-apply the protocol — which is precisely the omission that produced the one high-severity bug found in review.
• Built from a {attributes, stride, buffer, indexBuffer} descriptor — a GPUVertexBufferLayout + arrayStride — so [WebGPU port] Backend-neutral vertex layout descriptor #1492 becomes a class swap rather than an extraction.
• A batcher switch is one bindVertexArray; steady-state frames issue zero attribute-specification calls (19 vertexAttribPointer + 19 enableVertexAttribArray at startup, 0 per frame after; previously 3–5 of each per switch plus 3–4 per mesh flush).
• The attribute-leak machinery is gone — state is swapped wholesale, not disabled. That class of cross-batcher leak is now impossible by construction.
• Custom shaders hosted by a built-in batcher must declare that batcher's attributes first, in layout order (ShaderEffect-generated vertex shaders already comply); a console warning fires once per shader on mismatch.
• TMX GPU tilemap eligibility now uses renderer.supportsShaderTileLayers, a backend capability flag on the base Renderer, instead of a WebGL-version check.

Performance, honestly

Measured on the bundled examples, this removes ~2 GL calls/frame in a single-batcher scene and ~40 in one mixing sprites, meshes and primitives — well under a millisecond either way. The structural benefits (impossible state leaks, pipeline-shaped layout) are the real payoff; this is not an FPS change.

Verification

• 4888 tests passing (172 files, 1 skipped). 6 new specs: webgl_vertexstate (13 direct unit tests on the new class), plus VAO state / call-counts / recreation / contract / adversarial / teardown. The old batcher_attribute_leak spec is retired and its invariant re-expressed for the VAO era.
• Review pass by an independent agent found one high-severity bug (build steps leaked the global ARRAY_BUFFER binding, so rebuilding a non-current batcher corrupted the current one's uploads — silently, with NO_ERROR). Fixed at the root, with a regression test. Also fixed: VAOs were never released on destroy (plus pre-existing buffer/shader leaks), spurious warnings during lost-context re-init, and a swallowed error in renderer negotiation.
• Visual: 7 examples captured before/after. Four are pixel-identical (gltf, sprite-illuminator, mesh-3d-material, shader-effects); the three animated ones differ by less than their measured same-code capture noise, and were inspected directly.

Closes #1509

🤖 Generated with Claude Code

https://claude.ai/code/session_01JvxaXQRgQn5AoR8DNkv6Wg

[Copilot]

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

[Copilot]

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