parser: accept accessor keyword under experimentalDecorators: true

[robobun]

What

Fixes #29553. Fixes #27335. The accessor keyword (TC39 auto-accessors, TypeScript 4.9+) was gated behind the internal standard_decorators feature flag in the parser, so any class field written as public accessor foo: number failed to parse when a project's tsconfig.json had experimentalDecorators: true — a common default, especially in Angular-style projects.

# Reproduced at test/TEST1.ts:2 with v1.3.13-canary.1 (bf2e2cec)
2 |     public accessor computed: number;
                        ^
error: Expected ";" but found "computed"

Why

TypeScript accepts accessor in both decorator modes — they're independent proposals. The lowering path in lowerDecorators.zig already handles undecorated auto_accessor properties (WeakMap + getter/setter pair). All that was missing was the parser gate and the should_lower_standard_decorators condition.

How

• src/ast/parseProperty.zig: drop the p.options.features.standard_decorators guard on the p_accessor keyword branch. Always recognise accessor in class bodies.
• src/ast/parse.zig: set should_lower_standard_decorators whenever an auto-accessor is present (not only when standard_decorators is already on), so the accessor gets lowered correctly. JavaScriptCore doesn't parse accessor natively, so the rewrite is what makes it run.
• src/ast/parse.zig: if a class mixes accessor with TS legacy @dec members under experimentalDecorators: true, emit a clear compile error. Otherwise the legacy decorators would be routed through the standard-proposal runtime (wrong signature), which is worse than a clear error.

Verification

test/regression/issue/29553.test.ts — 5 tests covering:

• The issue's exact repro (bun test ./test with experimentalDecorators: true).
• accessor with all TS modifiers (public, private, protected, static, readonly) under experimentalDecorators: true.
• accessor in a plain TS file (no tsconfig).
• accessor under standard decorators (regression guard).
• The new error for legacy-@dec + accessor combined.

Before-fix: tests fail with the Expected ";" parse error. After-fix: 5 pass.

Nearby decorator regression suites (test/bundler/transpiler/es-decorators.test.ts, decorators.test.ts, decorator-metadata.test.ts, issue/27526.test.ts, issue/27575.test.ts) all still pass — 209 tests across those files, no regressions.

Related (not closed by this PR)

• Decorators on accessor class fields fail to parse in Bun #29197 (@decorator accessor x parse error) — this PR fixes it under the default (standard-decorators) mode, but under experimentalDecorators: true it now raises a clear "can't mix" error instead of silently producing wrong decorator semantics. Fully supporting legacy @dec + accessor in the same class requires a separate lowering pass (desugar accessor to #storage + getter/setter before the legacy-decorator loop) — out of scope here.
• this.#field not rewritten in class field initializers when class has @decorated accessor #28118 (this.#field not rewritten in initializers under @decorated accessor) — unrelated private-field rewrite bug in the decorator-lowering path; still reproduces after this PR.

[robobun]

^{Updated 2:11 PM PT - Apr 21st, 2026}

❌ @autofix-ci[bot], your commit b55e4dd has 2 failures in Build #46875 (All Failures):

• test/js/bun/webview/webview-chrome.test.ts - code 1 on 🍎 14 aarch64
• test/cli/install/bun-install.test.ts - code 1 on 🐧 13 x64-asan

---

🧪   To try this PR locally:

bunx bun-pr 29554

That installs a local version of the PR into your bun-29554 executable, so you can run:

bun-29554 --bun

[github-actions]

Found 3 issues this PR may fix:

1. Decorators on accessor class fields fail to parse in Bun #29197 - Decorators on accessor class fields fail to parse (@example accessor x produces Expected ";" but found "x")
2. TypeScript accessor keyword in classes fails to parse #27335 - TypeScript accessor keyword in classes fails to parse (public accessor name: string errors out)
3. this.#field not rewritten in class field initializers when class has @decorated accessor #28118 - this.#field not rewritten in class field initializers when class has @decorated accessor (related to should_lower_standard_decorators not being set for auto-accessor classes)

If this is helpful, copy the block below into the PR description to auto-close these issues on merge.

Fixes #29197
Fixes #27335
Fixes #28118

🤖 Generated with Claude Code

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro

Run ID: 5f872a85-51b2-4156-a8f2-71d46ff45c2e

📥 Commits

Reviewing files that changed from the base of the PR and between 131fecc and b55e4dd.

📒 Files selected for processing (1)

• test/regression/issue/29553.test.ts

---

Walkthrough

Parser recognizes accessor as an auto-accessor inside classes regardless of decorator mode, errors when auto-accessors coexist with legacy experimentalDecorators and any decorators, and makes auto-accessors follow the standard-decorator lowering path. A regression test exercises these cases.

Changes

Cohort / File(s)	Summary
Parser: class parsing src/ast/parse.zig	Emit an error if a class contains auto-accessor fields (accessor) while legacy decorator mode is active (!standard_decorators) and the class has any decorators. Update G.Class.should_lower_standard_decorators: auto-accessors force the standard lowering path; otherwise compute as standard_decorators && has_any_decorators.
Parser: property parsing src/ast/parseProperty.zig	Stop gating recognition of the accessor modifier on p.options.features.standard_decorators; treat accessor as .auto_accessor for class properties when opts.is_class and the modifier mapping applies (preserving raw/p_static mapping).
Tests: regression test/regression/issue/29553.test.ts	Add tests covering accessor auto-accessor parsing and runtime behavior across decorator configurations: with experimentalDecorators:true, without tsconfig, with experimentalDecorators:false plus a stage‑3 decorator, and an expected failure when mixing accessor with legacy decorators under experimentalDecorators:true.

🚥 Pre-merge checks | ✅ 4

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title accurately describes the main change: enabling the accessor keyword parser support under experimentalDecorators: true mode, which is the primary objective of the PR.
Description check	✅ Passed	The description covers both required sections: 'What' explains the problem and fix, 'How' details the implementation approach, and includes verification details with test coverage.
Linked Issues check	✅ Passed	The PR addresses all coding requirements from the linked issues [#29553, #27335]: parser now accepts accessor keyword in both decorator modes, auto-accessors are properly lowered, and the error handling for mixed legacy decorators is implemented with comprehensive test coverage.
Out of Scope Changes check	✅ Passed	All changes are directly scoped to the objectives: parser modifications in parseProperty.zig and parse.zig, error handling for decorator mixing, and regression tests. No unrelated changes detected.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[github-actions]

This PR may be a duplicate of:

1. Parse and lower accessor fields under experimentalDecorators #29201 - Same parser fix in parseProperty.zig and parse.zig to accept and lower accessor under experimentalDecorators: true; fixes the identical symptom (Decorators on accessor class fields fail to parse in Bun #29197 / TypeScript accessor keyword in classes fails to parse #27335).
2. feat(transpiler): lower auto-accessor class fields #26431 - Adds the same accessor parsing branch in parseProperty.zig and a lowerAutoAccessors pass; fixes the same underlying feature gap (Support TypeScript 4.9 class auto-accessors ("accessor" keyword) #6051).

🤖 Generated with Claude Code

[robobun]

Thanks — this is related to but not the same as #29201 / #26431.

• Parse and lower accessor fields under experimentalDecorators #29201 (10 days old, `mergeStateStatus: BLOCKED`, open CodeRabbit feedback): desugars `accessor` to `#storage` + `get`/`set` before the TS legacy decorator loop, handling the mixed `@legacyDec method` + `accessor x` case properly (~400 lines in `P.zig` + `visitExpr.zig`).
• feat(transpiler): lower auto-accessor class fields #26431 (3 months old, conflicting): Jarred's earlier take, different lowering location.
• This PR (parser: accept accessor keyword under experimentalDecorators: true #29554): minimal fix — drop the parser gate and route auto-accessor classes through the existing standard-decorator lowering. For the rare legacy-`@dec` + `accessor` combination, emit a clear compile error rather than silently applying standard-proposal decorator semantics to a legacy decorator. ~200 lines including tests.

If the comprehensive path in #29201 lands first, this PR should be closed. If #29201 stays stuck, this one unblocks the reported issues (#29553 and #27335) while leaving the mixed case as a surfaced "not supported" error instead of a silent miscompile.

[claude]

LGTM — the fix correctly threads accessor lowering through the existing WeakMap+getter/setter path by setting should_lower_standard_decorators whenever has_auto_accessor is true, regardless of decorator mode.

Extended reasoning...

Overview

Two small Zig changes and a new regression test file. parseProperty.zig drops the standard_decorators feature guard on the p_accessor branch so the keyword is accepted in class bodies under either decorator mode. parse.zig adjusts the should_lower_standard_decorators condition to has_auto_accessor or (standard_decorators and has_any_decorators) and adds a diagnostic for the mixed legacy-decorator + accessor case.

Security risks

None. This is a parser fix with no auth, crypto, or permissions code involved.

Level of scrutiny

Low-to-medium. The change is small and self-contained: one boolean condition in the property parser, one boolean condition in the class-body assembler, and one new error path. The existing lowerDecorators.zig already has the undecorated auto-accessor lowering path (WeakMap + getter/setter, line ~820), so no new lowering logic is introduced — the fix just ensures the right code path is reached. Five targeted regression tests cover the main scenario, modifier variants, no-tsconfig baseline, standard-decorator regression guard, and the new error case.

Other factors

The PR description notes that 209 tests across nearby decorator suites continue to pass. The mixed-mode error is a conservative safety net: rather than silently mis-route legacy decorators through the standard-proposal runtime, it rejects the combination with a clear diagnostic. That tradeoff is appropriate.

[robobun]

CI failures on b55e4ddf are pre-existing / infra, not caused by this PR:

• `debian-13-x64-asan`: test/cli/install/bun-install.test.ts → flaky InvalidMatcherError from toBeValidBin/toHaveWorkspaceLink. Same failure on unrelated PR sql: serialize JS arrays in sql(object) as PostgreSQL array literals #29552 at build #46873 — triggered by the concurrent-describe layout in that file interacting with the recent install refactor in install: global virtual store for isolated linker (7x faster warm installs) #29489. The four tests from that file ran locally against this PR (pass).
• `darwin-13-aarch64` / `darwin-14-aarch64`: Expired — buildkite agents never picked up the job. Pure infra timeout.

The regression test added here (test/regression/issue/29553.test.ts) passes under debug/ASAN locally — 4/4, and fails 2/4 with src/ reverted, so the gate holds. Nothing code-side to do on my end; happy to re-push / re-run if a maintainer kicks CI.

[alii]

@robobun Did you see #29201? How does it compare?

[robobun]

Yep — #29201 is the more comprehensive fix; this is a minimal subset. Breakdown:

#29201 (my earlier branch, 807+/10-, 4 files, 10 days old, BLOCKED):

• Parser: drop the gate (same as here).
• src/ast/P.zig: rewriteAutoAccessorProperties desugars every accessor field into #storage + get/set before the TS legacy decorator loop, so @legacyDec accessor x works end-to-end under experimentalDecorators: true. Handles decorated accessors, computed keys (single-eval via hoisted temp), private-identifier keys, static accessors, class-expression accessors, collision-avoiding backing names.
• src/ast/visitExpr.zig: runs the same rewrite on class expressions.
• Open CodeRabbit review feedback (4 items, mostly bun.handleOom style + one real concern about computed-key single-eval in class-expression fallback).

#29554 (this PR, 116+/3-, 3 files):

• Parser: drop the gate.
• Route auto-accessor classes through the existing standard-decorator lowering in lowerDecorators.zig (which already handles the undecorated accessor → WeakMap + get/set path at line 820). No new lowering code.
• For @legacyDec method() + accessor x in the same class: emit a clean compile error rather than silently mis-route the legacy decorator through the standard-proposal runtime.

Trade-off is narrow: #29553 and #27335 fully fixed by either, but #29201 supports decorated accessors under experimentalDecorators while this PR rejects that combination with a clear diagnostic.

Happy with either landing. If you want the comprehensive one, #29201 needs a rebase onto current main (branch is 10d old) and the CodeRabbit threads addressed — I can do that on top of this. If you want the minimal path forward for now, this is ready and gate-verified.

[alii]

Should we close this? What is more correct?

[robobun]

Honest take: #29201 is more correct, close this.

TypeScript compiles @legacyDec accessor x under experimentalDecorators: true by desugaring accessor to #storage + get/set first, then emitting __decorate([legacyDec], Foo.prototype, "x", null) on the synthesized getter. #29201 mirrors that model exactly. This PR rejects the same code with a clean error — a strictly narrower behavior than TS offers.

For the issue at hand (#29553), both work. For #29197 (@dec accessor x), only #29201 works.

I’ll rebase #29201 onto current main, address the 4 open CodeRabbit threads (mostly catch bun.outOfMemory → bun.handleOom, plus the computed-key single-eval in class-expression fallback), and close this one.

[robobun]

Closing in favor of #29201 — rebased onto current main, squashed the 13 commits into one clean diff, reworded the commit message, and verified locally:

• Regression test (test/regression/issue/29197.test.ts) — 17/17 pass.
• Nearby decorator suites (decorators.test.ts, decorator-metadata.test.ts, es-decorators*.test.ts, issue/27526.test.ts, issue/27575.test.ts) — 207/207 pass, no regressions.
• Issue Bun test issue when run through scripts (accessor) #29553 reporter's exact repro (bun test ./test with experimentalDecorators: true) — passes.
• Mixed-case @legacyDec method() + @legacyDec accessor x — both decorators now fire with the correct descriptor shape (get/set for the accessor, value for the method), matching TS output.

#29201 now also closes #29553, #27335, #29197.