Add no-unused-array-method-return rule

[sindresorhus]

Fixes #741

[silverwind]

Critical: rule crashes on common code

rules/no-unused-array-method-return.js:251 — getVariableValue returns definition.node.init unconditionally. When a declarator has no initializer (uninitialized let, for…of, for…in, for (let x; …)), init is null. resolveReceiver then recurses with null and throws TypeError: Cannot read properties of null (reading 'type') at line 171.

Reproduced via npx ava test/no-unused-array-method-return.js after dropping these inputs into the existing test file:

'let array; array.map(fn);'                    // CRASH
'for (const x of arr) x.map(fn);'              // CRASH
'for (let x of arr) x.map(fn);'                // CRASH
'for (var x of arr) x.map(fn);'                // CRASH
'for (const x in obj) x.map(fn);'              // CRASH
'for (let x; x.length;) x.map(fn);'            // CRASH

Suggested fix: add && definition.node.init to the condition at lines 244–250, returning uncertainValue for these cases (matches the documented "for…of bindings unsupported" intent). Worth adding at least one for…of and one bare let x; to the test file to prevent regression.

Code quality

• L9–40 — methodNames is only used to build methods; inline as const methods = new Set([...]);.
• L224–238 and L289–304 — ~30 lines of inline boundary commentary largely repeat what the docs already explain. Trimming would match the terser style of nearby rules.
• Mix of const x = (…) => and function x(…) declarations in the same file is inconsistent.

Heuristic gaps

• .slice / .at / .includes / .indexOf / .concat / .lastIndexOf exist on strings; unresolved receivers will false-positive. Acknowledged in the docs, but worth a final risk check before promoting beyond unopinionated.
• Statement-level SequenceExpression (e.g. sideEffect(), arr.map(fn);) is silently skipped — listed as intentional but a real false negative.

---

Review by Claude Opus 4.7