fix: rules-of-hooks FP on forwardRef/memo callbacks under non-PascalCase bindings

[devin-ai-integration]

Summary

rules-of-hooks flagged hooks inside React HoC render callbacks whenever the resulting binding wasn't PascalCase — the exact shape compiled/generated mobile code uses:

const _Wrapped = forwardRef((props, ref) => {
  useHook(); // FP: "`useHook` runs inside `_Wrapped`, which is not a component or Hook"
  ...
});

The component-or-hook decision keyed only on the inferred binding name (inferFunctionName sees through forwardRef/memo to _Wrapped, and isReactComponentName rejects _-prefixed names). But the callback passed directly to memo(...) / forwardRef(...) (incl. React.* forms, and nested memo(forwardRef(...))) is a component by construction, regardless of the binding name — same stance as upstream eslint-plugin-react-hooks' isForwardRefCallback/isMemoCallback.

 findEnclosingFunctionInfo:
+  isHocCallback = isReactHocCallbackArgument(fn)  // direct arg of REACT_HOC_NAMES call
-  isComponentOrHook: isReactComponentOrHookName(resolvedName)
+  isComponentOrHook: isHocCallback || isReactComponentOrHookName(resolvedName)

findEnclosingComponentOrHookFunction (useEffectEvent same-scope check) gets the same treatment. Arbitrary non-React wrappers (trackEvents(fn)) are NOT promoted, so a genuinely non-component _helper() calling hooks still fires.

Eval results

Full oxlint-plugin-react-doctor suite: 39 failures on this branch — byte-identical to the 39 preexisting failures on main (no new misses on the adversarial suite). New regression tests in rules-of-hooks.regressions.test.ts cover the forwardRef/memo/React.forwardRef/nested shapes plus both still-fires cases.

Test plan

• vp test run src/plugin/rules/react-builtins/rules-of-hooks*.test.ts (167 passed)
• pnpm --filter oxlint-plugin-react-doctor typecheck
• pnpm lint

Link to Devin session: https://app.devin.ai/sessions/9f543a0266b74767b52eb6b246c652d4
Requested by: @aidenybai

---

Note

Low Risk
Lint-rule heuristic change with targeted regression tests; no runtime or auth/data impact, and real violations (memo comparator, non-React HoCs) remain covered.

Overview
Fixes false positives in rules-of-hooks and exhaustive-deps when hooks run inside forwardRef / memo render callbacks assigned to non-PascalCase names (e.g. const _Wrapped = forwardRef(...)).

Adds isReactHocCallbackArgument, which treats the first argument to known React HoCs (memo, forwardRef, including React.*) as a component by construction. rules-of-hooks uses it in findEnclosingFunctionInfo and findEnclosingComponentOrHookFunction, and tightens the non-component branch so HoC callbacks are not misclassified via inferred binding names. exhaustive-deps uses the same helper so factory-scope captures (e.g. logger) are not reported as missing deps when the component boundary was wrong.

memo’s second-argument comparator, arbitrary wrappers, and plain _helper functions are unchanged and still report. New regression tests cover the fixed and still-firing cases.

^{Reviewed by Cursor Bugbot for commit 480f578. Bugbot is set up for automated code reviews on this repo. Configure here.}

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring

[pkg-pr-new]

Open in StackBlitz

npm i https://pkg.pr.new/eslint-plugin-react-doctor@794

npm i https://pkg.pr.new/oxlint-plugin-react-doctor@794

npm i https://pkg.pr.new/react-doctor@794

commit: 480f578

[github-actions]

No React Doctor issues found. 🎉

_{Reviewed by React Doctor for commit 480f578.}

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit f9849a5. Configure here.}

[devin-ai-integration]

Test results — built the react-doctor CLI from this branch and from clean main (d48e7f1) and diffed every rules-of-hooks diagnostic.

Escalation: I could not locate the RDE harness (OSS recording-repo manifest from .agents/skills/rule-validate/SKILL.md) on this machine, so the literal "29 known FPs → 0" RDE run was substituted with a deterministic E2E CLI run on the exact reported shape plus main-vs-branch CLI scans of 3 OSS RN repos. Happy to run the official flow if pointed at the harness.

• Exact reported shape E2E (fixture project, full CLI --json run): passed
• OSS no-regression scan (paper, bottom-sheet, gesture-handler): passed — diagnostics byte-identical main vs branch
• Regression: full plugin unit suite parity: passed — same 39 preexisting failures as main, +6 new passing regression tests

Test 1 — exact reported shape (CLI E2E)

Fixture src/wrapped.tsx: const _Wrapped = forwardRef((props, ref) => { useImperativeHandle(...); useState(...) }), const _Memoized = memo(...) with a hook, and non-component const _helper = () => { useState(...) }.

main CLI:

src/wrapped.tsx:4  useImperativeHandle inside `_Wrapped` … not a component or Hook   (FP)
src/wrapped.tsx:5  useState inside `_Wrapped` …                                      (FP)
src/wrapped.tsx:15 useState inside `_Memoized` …                                     (FP)
src/wrapped.tsx:25 useState inside `_helper` …                                       (true positive)

branch CLI:

src/wrapped.tsx:25 useState inside `_helper` …                                       (true positive kept)

Zero diagnostics on the HoC callbacks; _helper still fires.

Test 2 — OSS no-regression scan

Repo	main roh diags	branch roh diags	diff
callstack/react-native-paper	1	1	identical
gorhom/react-native-bottom-sheet	2	2	identical
software-mansion/react-native-gesture-handler	10	10	identical

No branch-only diagnostics, no disappearances. None of these repos contain the non-PascalCase HoC-binding shape in source (it's typical of compiled/generated bundles), hence the fixture in Test 1 carries the FP→0 proof; facebook/react-native source was also grep-verified to have no occurrences.

Test 3 — Regression: unit suite parity

Full oxlint-plugin-react-doctor suite: branch 39 failed | 6820 passed vs clean main 39 failed | 6814 passed; failure lists byte-identical (timing-only diffs). The +6 passes are the new tests in rules-of-hooks.regressions.test.ts.

CI: 18 passed, 0 failed. Devin session

[aidenybai]

Pushed review fixes in 480f578:

• memo comparator regression (also flagged by Bugbot): isReactHocCallbackArgument matched any argument position, so hooks inside memo(Component, comparator) — flagged before this branch — went silent. Promotion is now restricted to the first argument (the render callback), with a regression test asserting the exact diagnostic.
• hasResolvedName semantics: the field's doc comment ("true iff a name was actually inferred") no longer matched the || isHocCallback overload. The consumer now checks isComponentOrHook before the anonymous walk-out, which deletes the overload entirely. Behavior is unchanged for all combinations.
• Shared util + exhaustive-deps: the same FP existed in exhaustive-deps' copy of this machinery — factory-scope captures inside const _Wrapped = forwardRef(...) were reported as missing deps (confirmed with a failing test before the fix). isReactHocCallbackArgument now lives in utils/is-react-hoc-callback-argument.ts and both rules use it.
• Test strength: the two "still flags" tests now assert the exact diagnostic message instead of length > 0; the changeset covers the comparator restriction and the exhaustive-deps fix.

Verification: all 787 rules-of-hooks + exhaustive-deps tests pass; the full plugin suite's 39 failures are byte-identical to this branch's pre-fix baseline (same preexisting set); typecheck, lint, and format:check pass.