fix: reject unsigned unary negation

[huytdps13400]

Summary

• reject unary negation when the operand primitive is u32
• validate in neg.signature so the error is raised before both comptime folding and WGSL emission
• cover scalar unary - and unsigned-vector std.neg paths with exact resolution errors

WGSL does not define unary negation for unsigned integer scalars or vectors. This follows the maintainer direction in #2846 and avoids an implicit signed cast.

TDD evidence

RED: the scalar regression failed because tgpu.resolve() emitted code instead of throwing.

GREEN: the focused WgslGenerator suite passes 90/90 and both u32 and vec2u now fail at resolution with a descriptive WgslTypeError.

Verification

• package typecheck passes
• changed files pass oxlint and oxfmt
• source TypeGPU suite: 176 files, 2,364 tests pass
• built TypeGPU suite: 144 files, 1,941 tests pass
• circular dependency check passes
• git diff --check

Fixes #2846

[pullfrog]

✅ No new issues found.

Reviewed changes

• std/operators.ts — reject unsigned neg — neg.signature now throws WgslTypeError when getPrimitive(arg) === u32, so unary minus on unsigned scalars and vectors fails at resolution (before comptime folding and WGSL emission) instead of producing an implicit unsigned cast.
• wgslGenerator.test.ts — regression coverage — adds two exact toThrowErrorMatchingInlineSnapshot cases covering scalar - on u32 and std.neg on vec2u.

The fix is well-targeted: both syntax paths (-x via wgslGenerator.ts:172 and the std.neg export) dispatch through neg[$gpuCallable], so they all reach the new signature guard. getPrimitive unwraps vector/matrix composites to the singleton u32, making the identity check correct for vectors, and the message style matches the existing Unary operator ! error. The snapshots are exact-match and fail cleanly without the change, so the regression coverage is genuine.

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

[Copilot]

Pull request overview

This PR tightens TGSL/WGSL type resolution to reject unary negation (-) on unsigned integer operands (per WGSL rules), ensuring the error is raised during resolution (before comptime folding and WGSL emission), and adds regression tests for both scalar - and std.neg paths.

Changes:

• Add early validation in std.neg’s signature to throw a WgslTypeError for unsupported unsigned negation.
• Add WgslGenerator-focused tests asserting exact resolution error output for u32 scalar unary - and vec2u passed to std.neg.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
packages/typegpu/src/std/operators.ts	Adds signature-time validation for neg to reject unsigned operands with a WgslTypeError.
packages/typegpu/tests/tgsl/wgslGenerator.test.ts	Adds regression tests that assert resolution fails for -u32(...) and std.neg(vec2u).

---

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

[huytdps13400]

Thanks for flagging this. I traced both paths on the current head and they are already rejected before invalid WGSL can be emitted.

For u16, a focused resolver regression reaches the existing schema guard first and fails with 'u16 has no representation in WGSL'; it never reaches neg.signature. For bool and matrices, the public std.neg overloads come from cpuNeg(number | AnyNumericVecInstance), and TypeScript rejects both calls with TS2769. The signature callback is correspondingly typed through MapValueToDataType<Parameters>. Only an explicit unsafe cast can bypass that public contract.

The supported unsigned scalar/vector paths are u32 and vecNu; getPrimitive maps both to the same u32 singleton handled by this patch. I reran packages/typegpu/tests/tgsl/wgslGenerator.test.ts after the probes: 90/90 tests pass and the branch remains unchanged. I am keeping the fix scoped to the reachable unsigned WGSL operands from #2846.