fix: Restrict using slot for strings

[aleksanderkatan]

Includes changes from #2703 by @iwoplaza

[github-actions]

Bundle size comparison (import * as ... in PR vs import * as ... in target):

🟢 Decreased (max -0.20%)	➖ Unchanged	🔴 Increased (max 0.21%)	❔ Unknown
262	36	23	0

import { ... } in PR vs import * as ... in PR (is the library tree-Shakeable?):

Test	tsdown
tgpu_init.ts	256.72 kB ($${\color{green}-3.2\%}$$)
tgpu_initFromDevice.ts	256.21 kB ($${\color{green}-3.4\%}$$)
tgpu_resolve.ts	187.44 kB ($${\color{green}-29.3\%}$$)
tgpu_resolveWithContext.ts	187.37 kB ($${\color{green}-29.3\%}$$)
tgpu_bindGroupLayout.ts	68.97 kB ($${\color{green}-74.0\%}$$)
tgpu_mutableAccessor.ts	65.90 kB ($${\color{green}-75.1\%}$$)
tgpu_accessor.ts	65.88 kB ($${\color{green}-75.2\%}$$)
tgpu_privateVar.ts	65.26 kB ($${\color{green}-75.4\%}$$)
tgpu_workgroupVar.ts	65.26 kB ($${\color{green}-75.4\%}$$)
tgpu_const.ts	64.49 kB ($${\color{green}-75.7\%}$$)
tgpu_fn.ts	38.06 kB ($${\color{green}-85.6\%}$$)
tgpu_fragmentFn.ts	37.99 kB ($${\color{green}-85.7\%}$$)
tgpu_vertexFn.ts	37.81 kB ($${\color{green}-85.7\%}$$)
tgpu_computeFn.ts	37.51 kB ($${\color{green}-85.9\%}$$)
tgpu_vertexLayout.ts	26.60 kB ($${\color{green}-90.0\%}$$)
tgpu_comptime.ts	14.40 kB ($${\color{green}-94.6\%}$$)
tgpu_unroll.ts	2.59 kB ($${\color{green}-99.0\%}$$)
tgpu_slot.ts	1.57 kB ($${\color{green}-99.4\%}$$)
tgpu_lazy.ts	1.22 kB ($${\color{green}-99.5\%}$$)

If you wish to run a comparison for other, slower bundlers, run the 'Tree-shake test' from the GitHub Actions menu.

[github-actions]

Resolution Time Benchmark

---
config:
  themeVariables:
    xyChart:
      plotColorPalette: "#E63946, #3B82F6, #059669"
---
xychart
  title "Random Branching (🔴 PR | 🔵 main | 🟢 release)"
  x-axis "max depth" [1, 2, 3, 4, 5, 6, 7, 8]
  y-axis "time (ms)"
  line [0.69, 1.33, 2.65, 4.26, 5.15, 7.37, 13.87, 15.06]
  line [0.67, 1.33, 2.90, 4.29, 4.61, 7.40, 14.18, 17.30]
  line [0.70, 1.33, 2.78, 3.80, 4.64, 7.47, 13.82, 15.07]

Loading

---
config:
  themeVariables:
    xyChart:
      plotColorPalette: "#E63946, #3B82F6, #059669"
---
xychart
  title "Linear Recursion (🔴 PR | 🔵 main | 🟢 release)"
  x-axis "max depth" [1, 2, 3, 4, 5, 6, 7, 8]
  y-axis "time (ms)"
  line [0.20, 0.34, 0.46, 0.54, 0.77, 0.76, 0.92, 1.06]
  line [0.21, 0.46, 0.50, 0.62, 0.83, 0.87, 1.00, 1.09]
  line [0.20, 0.39, 0.45, 0.57, 0.74, 0.81, 1.02, 1.07]

Loading

---
config:
  themeVariables:
    xyChart:
      plotColorPalette: "#E63946, #3B82F6, #059669"
---
xychart
  title "Full Tree (🔴 PR | 🔵 main | 🟢 release)"
  x-axis "max depth" [1, 2, 3, 4, 5, 6, 7, 8]
  y-axis "time (ms)"
  line [0.73, 1.58, 2.66, 4.15, 8.14, 17.05, 36.32, 74.97]
  line [0.69, 1.48, 3.15, 4.26, 8.02, 16.86, 36.96, 75.77]
  line [0.76, 1.57, 3.35, 4.08, 8.18, 16.72, 36.58, 73.72]

Loading

[aleksanderkatan]

@pullfrog review

[pullfrog]

✅ No new issues found.

Reviewed changes — This PR prevents raw WGSL string injection through external maps, slots and TGSL expressions by requiring strings to carry a known schema type or be wrapped in a dedicated ResolvableString marker used for argument names.

• Adds ResolvableString to wrap raw-WGSL argument identifiers used during entry-function header generation.
• Tightens ResolutionCtx.resolve so that a bare string without a concrete schema throws a clear error instead of being injected.
• Replaces ctx.resolve(value, dataType).value calls with ctx.resolveSnippet across generation code for consistency.
• Removes strings from the isWgsl / Wgsl type union, since they are no longer first-class resolvable values.
• Adds tests covering direct, indirect, slot-based and TGSL string-injection cases, and updates existing slot tests to use d.v3f values instead of WGSL string snippets.

  ^{｜ View workflow run ｜ Using Kimi K2 (free via Pullfrog for OSS) ｜ 𝕏}

[pullfrog]

✅ No new issues found.

Reviewed changes — This PR prevents raw WGSL string injection through external maps, slots and TGSL expressions by requiring strings to carry a known schema type or be wrapped in a dedicated ResolvableString marker used for argument names.

• Adds ResolvableString to wrap raw-WGSL argument identifiers used during entry-function header generation.
• Tightens ResolutionCtx.resolve so that a bare string without a concrete schema throws a clear error instead of being injected.
• Replaces ctx.resolve(value, dataType).value calls with ctx.resolveSnippet across generation code for consistency.
• Removes strings from the isWgsl / Wgsl type union, since they are no longer first-class resolvable values.
• Adds tests covering direct, indirect, slot-based and TGSL string-injection cases, and updates existing slot tests to use d.v3f values instead of WGSL string snippets.

  ^{｜ View workflow run ｜ Using Kimi K2 (free via Pullfrog for OSS) ｜ 𝕏}

[pullfrog]

Run failed. View the logs →

  ^{｜ Rerun failed job ➔ ｜ View workflow run ｜ via Pullfrog ｜ Using Kimi K2 (free via Pullfrog for OSS) ｜ 𝕏}

[pullfrog]

Important

The runtime guard now rejects strings, but the Wgsl type still advertises them. Tightening the type would keep string-injection errors compile-time instead of runtime.

Reviewed changes — restrict raw string injection into WGSL and route internal identifier strings through a new ResolvableString wrapper.

• Introduced internal ResolvableString so raw-function positional argument names can still be emitted as valid WGSL identifiers.
• Updated ResolutionCtxImpl.resolve to throw when a string is resolved without a concrete schema, blocking string injection via externals, slots, and TGSL identifiers.
• Converted generator/accessor sites from ctx.resolve(value, dataType) to ctx.resolveSnippet(snippet).
• Updated tests to stop relying on string-valued slots and added coverage for the new restrictions plus a comptime-string-slot test.

⚠️ Wgsl type/runtime mismatch

packages/typegpu/src/types.ts:75 still defines Wgsl as Eventual<string | number | boolean | ResolvableObject>, but isWgsl at types.ts:425 no longer accepts strings. The mismatch means users can pass a string to resolveWithContext(...).externals or the low-level resolve(item: Wgsl, ...) and have it type-check but throw at runtime. Removing string from the Wgsl type would make this restriction compile-time.

⚠️ String injection still possible with a concrete schema

packages/typegpu/src/resolutionCtx.ts:961 throws for strings only when no schema (or UnknownData) is supplied, but the branch at resolutionCtx.ts:968 still returns snip(item, schema, 'runtime') when a concrete schema is provided. This is an internal hole in the "no raw strings" rule. If no internal call site depends on it, the branch should be removed; otherwise it warrants a comment explaining the legitimate use case.

---

Unable to submit via the review API in this environment, so posting the review as a comment.

  ^{｜ View workflow run ｜ via Pullfrog ｜ Using Kimi K2 (free via Pullfrog for OSS) ｜ 𝕏}

[pullfrog]

Technical note: I attempted to submit the above feedback via create_pull_request_review multiple times, but the MCP tool consistently skipped because it did not receive the body/comments parameters from my calls. I was unable to get that tool to create an actual PR review in this run.

  ^{｜ View workflow run ｜ via Pullfrog ｜ Using Kimi K2 (free via Pullfrog for OSS) ｜ 𝕏}

[Copilot]

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

[pullfrog]

✅ No new issues found.

Reviewed changes — The latest commit aligns the type system with the runtime string-injection restriction.

• Removed string from Wgsl and isWgsl() so plain JS strings no longer type-check as resolvable values, matching the runtime rejection added earlier.
• Clarified the scoped string-identifer path in ResolutionCtx.resolve() with an inline example showing how identifier snippets carry their concrete schema.

  ^{｜ View workflow run ｜ Using Kimi K2 (free via Pullfrog for OSS) ｜ 𝕏}