[k7d]

Based on main.

In ES module output, sequential destructuring of a let local that certainly holds a JavaScript array reads it by index, (v?.[i] ?? null), instead of calling nth. The result is the same as nth's: the element within the array's bounds, nil past its end.

(let [[x y] #js [1 2]] ...)

const vec__1 = [1, 2];
const x = (vec__1?.[0] ?? null);
const y = (vec__1?.[1] ?? null);

What counts as an array: only a let local bound to a #js [] literal (cljs.analyzer/js-array-init?). Everything else goes through nth as before:

• loop locals and params, which recur can rebind to other values;
• values merely hinted ^array (a hinted param that gets a vector would break);
• other JS calls.

React's array-returning hooks (useState and the like) are added to js-array-init? in #17, with the rest of the React Compiler logic.

Performance

Micro-benchmark (release build, rolldown-minified, Node 26, 3 runs):

Benchmark	before	this branch
destructure a JS array, 10M times	~113 ms	~45 ms (2.5× faster)
destructure a vector, 10M times (unchanged path)	~66 ms	~63 ms

Tests

• script/test-esm: 0 failures, 0 errors. A new build test checks that #js [] literals are read by index (nil past their end), that vectors, other JS calls, loop locals and ^array hints go through nth, and that the output runs correctly.
• Compiler tests: 0 failures, 0 errors.

[k7d]

Fixes from review (fdba401):

• Past the end, v?.[i] was undefined, not nil. It now emits (v?.[i] ?? null), as nth returns.
• Loop locals were trusted to stay arrays. Only let locals qualify now; recur can't assign those.
• An ^array hint changed runtime semantics. A hinted param that received a vector got nil. Hints no longer qualify; only values certain to be arrays do.
• React hook names were matched by name in the compiler. cljs.analyzer/js-array-init? now decides, by module and fn: react's (and preact/hooks', preact/compat's) useState, useReducer, useTransition, useActionState, useOptimistic and useFormState, or a #js [] literal. Another module's useState goes through nth.

The test now covers each case. script/test-esm and the compiler tests pass.

[k7d]

Restacked on main with #21 merged (312dfa40), force-pushed:

• ES module output: const and let instead of var in functions #14 esm-const → main: 3 commits
• ES module output: for's iterators call themselves through a letfn local #15 esm-for-self-ref → esm-const: 2 commits
• ES module output: destructure #js [] arrays by index #16 esm-destructure-arrays → esm-for-self-ref: 2 commits
• React Compiler: the Vite plugin's reactCompiler option, "use memo" directives, a react-compiler template #17 react-compiler-template → esm-destructure-arrays: 1 commit, its changes squashed. #x (#jsx) literals: hiccup compiled to react/jsx-runtime calls #13's #x commits and the merge commits are gone, now that main has them.

The review-fix merges are folded into the branches' own commits. The code is unchanged apart from #21's changes. Only CHANGELOG.md and the build tests needed conflicts resolved, and each branch's [Next] section now has main's entries plus its own.

On every branch, script/test-esm and the compiler tests pass, including #21's new test-locals-munged-to-the-same-name. On #17 the Node tests pass too, and the react-compiler template works in the dev server (memoized rows).