Summary

Ships the Canvas 2D half of Tier-1 #8 of #446. This is the surface idaptik-ums (App.res: 1178 LoC of DOM + canvas) and every non-Pixi UI in the estate has been waiting for; WebGL / WebGL2 / WebGPU stays at ○ until a consumer surfaces a concrete need.

What lands

stdlib/Canvas.affine (+170 lines, new module): 1 extern type (Ctx2D) + 26 extern fns covering the full idaptik-ums-relevant Canvas 2D surface:

lib/codegen_deno.ml (+58 lines): 26 __as_canvas* prelude helpers + 26 dispatch entries adjacent to the Ipc block.

tests/codegen-deno/canvas_smoke.{affine,harness.mjs} (+160 lines combined): 5 distinct smoke functions exercise every shipped extern. The MockCtx2D records every method call as a typed-op tuple so the test asserts call order + arguments, not just side-effects. The canvasMeasureText round-trip exercises the Json-return shape (the TextMetrics-shaped object).

docs/bindings-roadmap.adoc row #8 status promoted ○ → ◑; deferred items captured (WebGL, gradients/patterns, ImageData, curve primitives, compositing/clip).

Design notes

Why is HTMLCanvasElement not its own extern type? The canvas-creation entry point is host-dependent — browser document.createElement("canvas") vs idaptik's pre-existing DOM tree vs jsdom-under-Deno. Treating the canvas as opaque Json defers the typed wrapper to the natural follow-up once affinescript-dom lands runtime support (currently blocked on the wasm-codegen for-in / while gap, issue #255).

Why fix arc counter-clockwise to false? The 6-arg shape (ccw boolean) would force every call site to pass a literal false. The 5-arg shape covers the overwhelming majority of consumers; a typed wrapper with the ccw flag is a follow-up if a consumer surfaces the need.

Why two separate drawImage externs instead of one with optional args? AffineScript doesn't have JS-style variadic functions; making w and h Option<Float> would force every natural-size call to thread Nones. The two-extern split keeps each call site simple.

Why open-string eventMode / textAlign / textBaseline? Same rationale as the PixiJS expansion (#502) — Pixi 8's eventMode and Canvas's text-alignment values are open enumerations, and a sum-type binding would either freeze the set or require tagged-variant codegen that doesn't exist on the Deno-ESM backend yet (deferred to json.affine v0.3). Consistency with existing patterns.

Test plan

• dune build bin/main.exe — clean (only the expected parser warnings)
• dune runtest --force — 356 tests pass (was 354 pre-PR; +2 from new Canvas + Ipc modules' AOT smoke)
• tools/run_codegen_deno_tests.sh — all 19 harnesses including the new canvas_smoke.harness.mjs OK
• CI build job
• CI tools/run_codegen_deno_tests.sh job
• CI governance + Hypatia (6 baselines per repo CLAUDE.md may be red — pre-existing on main, not regressions)

Refs

• Umbrella: [campaign] AffineScript top-50 framework bindings roadmap (UMBRELLA) #446 (Tier 1 — idaptik blockers)
• Tier-1 sub-issue: [#446 STEP 2] Tier 1 — idaptik blockers (10 bindings) #450
• Row updated: docs/bindings-roadmap.adoc row ci: Bump dtolnay/rust-toolchain from efa25f7f19611383d5b0ccf2d1c8914531636bf9 to 3c5f7ea28cd621ae0bf5283f0e981fb97b8a7af9 #8
• Pattern siblings shipped this session: feat(stdlib): PixiJS 8.x Container transform-and-event surface (bindings #1) #502 (PixiJS Container expansion), feat(stdlib): Ipc.affine — MessageChannel + structuredClone (bindings #9) #506 (Ipc.affine)
• Adjacent / future: affinescript-dom (gated on wasm codegen: for-in/while loop bodies never execute in compiled programs (pre-existing, untested) #255), WebGL/WebGPU (Tier 3 axis, not surfaced as need yet)

🤖 Generated with Claude Code