feat: Private property access

[aleksanderkatan]

Blocked by #2425

Only the case with both "#prop" and #prop present breaks.

  it('...', () => {
    const cls = new (class {
      '#const' = tgpu.const(d.u32, 0);
      #const = tgpu.const(d.u32, 1);

      fn = () => {
        'use gpu';
        const a = this['#const'].$;
        const b = this.#const.$;
      };
    })();

    expect(tgpu.resolve([cls.fn])).toMatchInlineSnapshot(`
      "const item: u32 = 0u;

      fn fn_1() {
        const a = item;
        const b = item;
      }"
    `);
  });

[github-actions]

pkg.pr.new

packages
Ready to be installed by your favorite package manager ⬇️

https://pkg.pr.new/software-mansion/TypeGPU/eslint-plugin-typegpu@a1c4c89450b7df107a86ca5e5791f74032bfae33

https://pkg.pr.new/software-mansion/TypeGPU/tgpu-gen@a1c4c89450b7df107a86ca5e5791f74032bfae33

https://pkg.pr.new/software-mansion/TypeGPU/tinyest-for-wgsl@a1c4c89450b7df107a86ca5e5791f74032bfae33

https://pkg.pr.new/software-mansion/TypeGPU/typegpu@a1c4c89450b7df107a86ca5e5791f74032bfae33

https://pkg.pr.new/software-mansion/TypeGPU/@typegpu/cli@a1c4c89450b7df107a86ca5e5791f74032bfae33

https://pkg.pr.new/software-mansion/TypeGPU/@typegpu/color@a1c4c89450b7df107a86ca5e5791f74032bfae33

https://pkg.pr.new/software-mansion/TypeGPU/@typegpu/gl@a1c4c89450b7df107a86ca5e5791f74032bfae33

https://pkg.pr.new/software-mansion/TypeGPU/@typegpu/noise@a1c4c89450b7df107a86ca5e5791f74032bfae33

https://pkg.pr.new/software-mansion/TypeGPU/@typegpu/radiance-cascades@a1c4c89450b7df107a86ca5e5791f74032bfae33

https://pkg.pr.new/software-mansion/TypeGPU/@typegpu/react@a1c4c89450b7df107a86ca5e5791f74032bfae33

https://pkg.pr.new/software-mansion/TypeGPU/@typegpu/sdf@a1c4c89450b7df107a86ca5e5791f74032bfae33

https://pkg.pr.new/software-mansion/TypeGPU/@typegpu/three@a1c4c89450b7df107a86ca5e5791f74032bfae33

https://pkg.pr.new/software-mansion/TypeGPU/unplugin-typegpu@a1c4c89450b7df107a86ca5e5791f74032bfae33

benchmark
view benchmark

commit
view commit

[github-actions]

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

🟢 Decreased	➖ Unchanged	🔴 Increased	❔ Unknown
0	321	0	0

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

Test	tsdown
tgpu_init.ts	255.87 kB ($${\color{green}-3.2\%}$$)
tgpu_initFromDevice.ts	255.36 kB ($${\color{green}-3.4\%}$$)
tgpu_resolve.ts	161.34 kB ($${\color{green}-38.9\%}$$)
tgpu_resolveWithContext.ts	161.28 kB ($${\color{green}-39.0\%}$$)
tgpu_bindGroupLayout.ts	68.92 kB ($${\color{green}-73.9\%}$$)
tgpu_mutableAccessor.ts	65.91 kB ($${\color{green}-75.1\%}$$)
tgpu_accessor.ts	65.90 kB ($${\color{green}-75.1\%}$$)
tgpu_privateVar.ts	65.27 kB ($${\color{green}-75.3\%}$$)
tgpu_workgroupVar.ts	65.27 kB ($${\color{green}-75.3\%}$$)
tgpu_const.ts	64.52 kB ($${\color{green}-75.6\%}$$)
tgpu_fn.ts	37.98 kB ($${\color{green}-85.6\%}$$)
tgpu_fragmentFn.ts	37.88 kB ($${\color{green}-85.7\%}$$)
tgpu_vertexFn.ts	37.70 kB ($${\color{green}-85.7\%}$$)
tgpu_computeFn.ts	37.40 kB ($${\color{green}-85.8\%}$$)
tgpu_vertexLayout.ts	26.63 kB ($${\color{green}-89.9\%}$$)
tgpu_comptime.ts	14.43 kB ($${\color{green}-94.5\%}$$)
tgpu_unroll.ts	2.60 kB ($${\color{green}-99.0\%}$$)
tgpu_slot.ts	1.54 kB ($${\color{green}-99.4\%}$$)
tgpu_lazy.ts	1.19 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.

[aleksanderkatan]

@pullfrog review

[pullfrog]

Important

The Rollup externals test isn't actually testing the Rollup path, and the combined public/private #prop case silently resolves the wrong resource. Both need addressing before merge.

Reviewed changes — this PR enables JavaScript private class properties as TypeGPU externals, adds auto-naming for ClassPrivateProperty, and removes the lint rule that unconditionally banned private identifiers.

• Lift the blanket PrivateIdentifier lint ban in packages/eslint-plugin/src/rules/noUnsupportedSyntax.ts so this.#prop access is allowed.
• Track #prop chains in packages/tinyest-for-wgsl/src/externals.ts by handling both Babel PrivateName and acorn PrivateIdentifier.
• Auto-name private class fields in packages/unplugin-typegpu/src/core/common.ts via a new ClassPrivateProperty visitor and matching extractLabelledExpression branch.
• Add test coverage for private property access across tinyest-for-wgsl, typegpu, and unplugin-typegpu.

Collision between private #prop and string-key "#prop"

A class that defines both a public '#prop' and a private #prop currently routes both this['#prop'].$ and this.#prop.$ through the same this.#prop external key. The runtime object-based external replacement then resolves the public property value, so the private field value is silently dropped. This matches the limitation noted in the PR body; please either fix the key encoding or add a tracking issue/TODO so the limitation is visible to users and future maintainers.

Technical details

# Collision between private `#prop` and string-key `"#prop"`

## Affected sites
- packages/tinyest-for-wgsl/src/externals.ts:32-35 — private names are encoded as `this.#prop`, identical to the runtime object-property replacement path used for `this["#prop"]`.
- packages/typegpu/src/core/resolve/externals.ts:104-129 — object external replacement uses `externalName\\.(?<prop>...)` and matches `this.#prop` for `this` externals.

## Required outcome
- `this.#prop` (private field) and `this["#prop"]` (public string key) resolve to distinct resources when both are present.

## Suggested approach
Consider encoding private names with a distinct prefix or namespace (e.g., `this.private.#prop`) and teaching the resolver to map that prefix back to the private-field accessor. Alternatively, document the limitation and add a failing/todo test linking to an issue.

## Open questions for the human
- Is the combined case expected to be fixed in this PR, or is there a follow-up issue already planned? Please link it in a code comment so the next agent doesn't re-discover it.

  ^{｜ Fix all ➔ ｜ Fix 👍s ➔ ｜ View workflow run ｜ Using Kimi K2 (free via Pullfrog for OSS) ｜ 𝕏}

[aleksanderkatan]

@reczkok @iwoplaza @cieplypolar this is reviewable again

[pullfrog]

\u2139\ufe0f No critical issues \u2014 minor suggestions inline.

Reviewed changes \u2014 Adds support for accessing class private properties (this.#prop) from 'use gpu' functions by teaching external-chain extraction and auto-naming to recognize private property AST nodes.

• Removes the PrivateIdentifier ban in noUnsupportedSyntax.
• Extends tryFindExternalChain to build chain strings like this.#prop.
• Extends unplugin-typegpu auto-naming to ClassPrivateProperty nodes and assignments to private properties.
• Adds parser, resolution, auto-naming, and nested-externals tests for private property access.

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

[pullfrog]

One small test-coverage suggestion: the PR body calls out the case where both a public string property named "#prop" (this['#prop']) and a private property #prop exist in the same class. The new tests cover each in isolation (packages/typegpu/tests/externalPropAccess.test.ts lines 95 and 135), but a single regression test with both present would guard the exact collision scenario this PR fixes.

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

[iwoplaza]

Looks great! 💜