perf(plugin): early-exit sweep — cheap discriminators before expensive walks

[rayhanadev]

Why

The multi-agent perf audit's largest cross-cutting theme: ~24 rules ran subtree walks, scope traces, parent climbs, or settings parses before a trivial name/type/substring check that usually fails. This sweep reorders them and adds whole-file no-op gates where a Program-level fact (no zod import, no recycler-list import, no JSX) makes the rule inert. Findings: #6 #11 #18 #19 #42 #56 #100 #121 #129 #193 #205 #206 #217 #236 #237 #238 #259 #299 #308(partial) #327 #385 #386 #387 #428.

Before / After: pure predicate reorders and gates — the reported diagnostic set is provably unchanged (every gate is an ANDed condition of the existing report path).

What changed

• a11y: raw-name bails before getElementType/generated-image climbs (alt-text, anchor-is-valid, img-redundant-alt, interactive-supports-focus, role-supports-aria-props) via a new has-jsx-a11y-settings util (safe: without settings the alias sets are empty).
• Whole-file gates: zod rules on a literal "zod" import (the detectors cannot fire without one — verified), react-native list rules on recycler package imports, rn-no-raw-text on JSX presence, nextjs-no-side-effect-in-get-handler on filename at Program.
• security-scan insecure-crypto-risk: trigger-token pre-filter over raw content (conservative: the stripper only blanks characters, so stripped-content tokens are always present raw).
• Assorted: render-name pattern before scope traces (no-render-in-render), value-shape checks before parent climbs (prefer-stable-empty-fallback), charCodeAt uppercase check + hoisted Set (no-unstable-nested-components), first-match early exits (containsFetchCall, is-create-element-call).
• Deliberately NOT done (finding overreach, catalog annotated): fix(effect-needs-cleanup): recognize AbortController.abort() as a valid cleanup #308's whole-file react-import gate would drop real diagnostics on global-React files — only the safe single-lookup half shipped.

Test plan

• Plugin suite: 8704 pass / 94 skip — zero deltas vs pristine main. Root typecheck + lint clean.
• Equivalence: 617-file corpus scan before/after — 291 diagnostics, sorted tuples byte-identical.
• Adversarial review: an independent reviewer attacked every reorder and gate (side effects, message-choice changes, gate-excluded firing paths) — no divergence found.

Note: touches interactive-supports-focus.ts and rn-no-raw-text.ts, which the set-membership sweep PR also touches — whichever merges second needs a trivial rebase.

🤖 Generated with Claude Code

---

Note

Low Risk
Changes are predicate reordering and conservative skip gates only; PR claims byte-identical corpus diagnostics, so behavioral risk is low aside from any missed edge case in gate logic.

Overview
Performance-only refactor across ~23 oxlint rules: cheap checks and whole-file gates run before AST walks, scope lookups, settings parsing, and regex-heavy className work. Reported diagnostics are intended to stay identical (gates are conservative AND conditions on existing report paths).

Whole-file gates at Program: Zod v4 rules skip files without a "zod" import; FlashList/LegendList rules skip without recycler package imports (RECYCLABLE_LIST_PACKAGE_SOURCES); rn-no-raw-text skips wrapper classification when the file has no JSX; nextjs-no-side-effect-in-get-handler hoists route filename checks once per file.

a11y: New hasJsxA11ySettings lets alt-text and anchor-is-valid bail on raw JSX tag names before getElementType. Other a11y rules reorder checks (e.g. role-supports-aria-props collects aria-* first; interactive-supports-focus requires role + handlers earlier).

Misc: Tailwind design rules add substring guards before regex; insecure-crypto-risk pre-filters raw content; containsFetchCall and is-create-element-call stop after first match; assorted predicate reordering (no-render-in-render, js-hoist-intl, prefer-stable-empty-fallback, no-unstable-nested-components, server-cache-with-object-literal, no-create-context-in-render import helpers).

^{Reviewed by Cursor Bugbot for commit 4f82a61. Bugbot is set up for automated code reviews on this repo. Configure here.}

[Copilot]

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

[pkg-pr-new]

Open in StackBlitz

npm i https://pkg.pr.new/eslint-plugin-react-doctor@1047

npm i https://pkg.pr.new/oxlint-plugin-react-doctor@1047

npm i https://pkg.pr.new/react-doctor@1047

commit: 4f82a61