impr: auto-interpolate integer varyings

[huytdps13400]

Summary

• automatically add @interpolate(flat) to integer scalar and vector varyings
• apply the behavior only to vertex outputs and fragment inputs, never integer builtins, vertex attributes, or fragment outputs
• preserve explicit interpolation and custom location attributes
• cover shelled and inferred render pipelines plus the integer-builtin boundary

Closes #2147.

Verification

• RED: focused render-pipeline regressions failed because implicit integer varyings had no flat interpolation
• pnpm test:fast-unit — 216 files, 2,795 tests passed; 2 baseline skips
• pnpm --filter typegpu test:types
• pnpm test:style — zero warnings/errors
• pnpm --filter typegpu build
• git diff --check

[Copilot]

Pull request overview

Automatically applies @interpolate(flat) to eligible integer vertex outputs and fragment inputs while preserving explicit attributes, locations, and builtins.

Changes:

• Adds integer scalar/vector interpolation handling.
• Supports typed and inferred render pipelines.
• Adds regression coverage and updates snapshots.

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
packages/typegpu/tests/renderPipeline.test.ts	Adds regression tests and updates snapshots.
packages/typegpu/src/data/autoStruct.ts	Propagates interpolation configuration.
packages/typegpu/src/core/function/tgpuVertexFn.ts	Enables interpolation for vertex outputs.
packages/typegpu/src/core/function/tgpuFragmentFn.ts	Enables interpolation for fragment inputs.
packages/typegpu/src/core/function/ioSchema.ts	Applies flat interpolation to eligible integer varyings.
packages/typegpu/src/core/function/autoIO.ts	Configures inferred pipeline I/O.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[pullfrog]

ℹ️ No critical issues — one test-coverage suggestion inline.

Reviewed changes

• Auto-flat interpolation for integer varyings — new withFlatInterpolationForInteger in ioSchema.ts, gated behind a new autoInterpolateIntegers flag threaded through withLocations/separateBuiltins/createIoSchema. Wraps only non-decorated or location-decorated integer scalars/vectors with @interpolate(flat), and is idempotent when an @interpolate attribute already exists.
• Scoped to vertex out / fragment in only — the flag is true only on the vertex-output and fragment-input paths (tgpuVertexFn.ts, tgpuFragmentFn.ts, and both AutoStruct paths in autoIO.ts). Vertex attribute inputs, fragment outputs, compute fns, and integer builtins keep the flag off.
• Tests — two new render-pipeline regressions (explicit struct layout and inferred varyings) plus updated snapshots pinning the emitted @interpolate(flat).

I traced the change through the count of reachable wiring sites and the emission path; the wiring matches the stated scope (tgpuVertexFn.ts:180, tgpuFragmentFn.ts:212, autoIO.ts:89/142). The createIoSchema(returnType) call in resolutionCtx.ts:767 (left at the default false) only fires for primitive returns — fragment color outputs / non-varying primitives — which the PR intentionally excludes, so no coverage gap there. 'flat' is a valid FlatInterpolationType, so auto-generated output is spec-legal.

  ^{｜ Fix all ➔ ｜ Fix 👍s ➔ ｜ View workflow run ｜ Using DeepSeek Flash (free via Pullfrog for OSS) ｜ 𝕏}

[pullfrog]

✅ No new issues found.

Reviewed changes

• Added the scope-locking regression the prior review requested — "does not flat interpolate integer vertex inputs or fragment outputs" now asserts a vertex in: integer is emitted with only its @location and a fragment out: integer keeps only its @location, with neither getting @interpolate(flat). This genuinely pins the boundary (a leak to those two paths would fail the assertions), closing the coverage gap previously called out. The earlier positive tests already covered the fragment-input/vertex-output side, so the four-position scope is now fully locked from both directions.

ℹ️ Nitpicks

• Test-only diff; nothing to nitpick. The explicit vertexFn/fragmentFn forms used here are the right complement to the existing inferred-form regressions.

  ^{｜ View workflow run ｜ Using DeepSeek Flash (free via Pullfrog for OSS) ｜ 𝕏}