fix(hir): materialize the implicit undefined for init-less let/const (#6871)

[proggeramlug]

Fixes #6871.

Problem

A let x; declared inside a loop body kept the previous iteration's value
instead of starting from undefined, whenever the assignment happened inside a
nested loop:

const src: string[][] = [["a"], [], ["b"]];
for (let i = 0; i < src.length; i++) {
  let f: string[] | undefined;
  for (const n of src[i]) { if (!f) f = []; f.push(n); }
  console.log(f);
}

	node	perry (before)
	["a"] / undefined / ["b"]	["a"] / ["a"] / ["a","b"]

Executing a lexical declaration initializes the binding to undefined
(ECMA-262 CreateMutableBinding + InitializeBinding), and re-executing it
creates a fresh binding — so every iteration must start from undefined.

Assignment placed directly in the loop body was already correct, and writing
= undefined explicitly was already correct. That combination is what kept
this hidden.

Cause

HIR faithfully emitted Stmt::Let { init: None }. Codegen allocates the slot
once in the entry block and emits no store for a None init, so the stale
value survived the loop back-edge.

This cannot be fixed in codegen without new information: Stmt::Let carries no
var/let distinction, and ctx.var_hoisted_ids never reaches perry-codegen.
Lowering is where is_var_decl is known, so the implicit initializer is
materialized there instead.

var is deliberately excluded — it is function-scoped and hoisted, so
re-executing var x; keeps the prior value.

Test

test-files/test_gap_uninit_let_loop_reset.ts — nested for-of, nested
while, nested for, scalar and string accumulators, an outer while, a
nested loop that never runs, plus the two forms that were already correct
(explicit = undefined, assignment directly in the body) to lock them in.
Byte-identical to node --experimental-strip-types.

Regression check

This touches every let declaration, so beyond the unit tests I ran a
differential sweep of 120 test_gap_* files against
node --experimental-strip-types: 114 pass, 2 fail, 4 skipped. Both
failures (test_gap_2159_defineproperty_class_prototype,
test_gap_2514_settracesigint) are already listed in
test-parity/known_failures.json.

cargo test -p perry-hir matches main exactly (18 passed / 1 failed —
logical_property_assignment_short_circuits_the_store_4586 fails identically
on origin/main, verified by reverting just this hunk).

Related

While testing I found the mirror-image defect in var: var x; in a loop is
reset each iteration when it should keep its value. That is pre-existing
(reproduces on released 0.5.1239, before this change) and is filed separately
as #6876. This PR neither fixes nor worsens it; the test documents the case in
a comment rather than asserting it, so no known failure gets baked in.

How this was found

Compiling the Milo compiler (https://github.com/milo-language/milo) with Perry.
Its parser collects declaration attributes with exactly this shape:

let attrs: Attribute[] | undefined;
while (this.at(TokenKind.At)) { if (!attrs) attrs = []; attrs.push(...); }

so a struct's @derive(...) leaked onto every following declaration and the
compiled compiler rejected valid programs ("'@derive' is not supported on
functions — 'arenaNew'"). Changing that one declaration to = undefined fixed
it, confirming the cause.

Summary by CodeRabbit

• Bug Fixes

  • Fixed uninitialized let variables retaining values from previous loop iterations.
  • Ensured let declarations reset to undefined correctly, including in nested loops.

• Tests

  • Added coverage for nested loop scenarios, uninitialized declarations, and reads before assignment.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: e3778bbd-24fb-4d87-a0fc-0a78aed8f933

📥 Commits

Reviewing files that changed from the base of the PR and between 0910cc7 and 1e6eaf9.

📒 Files selected for processing (3)

• changelog.d/6871-uninit-let-loop-reset.md
• crates/perry-hir/src/destructuring/var_decl.rs
• test-files/test_gap_uninit_let_loop_reset.ts

---

📝 Walkthrough

Walkthrough

The compiler now explicitly initializes uninitialized let and const declarations to undefined during lowering. Regression tests cover nested loops, assignments, reads, and non-executing inner loops.

Changes

Uninitialized let reset

Layer / File(s)	Summary
Materialize lexical declaration initializers crates/perry-hir/src/destructuring/var_decl.rs	Lowering emits undefined for uninitialized let and const declarations while preserving existing var behavior.
Validate nested-loop reset behavior test-files/test_gap_uninit_let_loop_reset.ts, changelog.d/6871-uninit-let-loop-reset.md	Regression cases cover nested loop variants, conditional assignment, reads before assignment, explicit undefined, and inner loops that do not execute; the changelog records the fix.

Estimated code review effort: 2 (Simple) | ~10 minutes

Possibly related issues

• #6876: Concerns the complementary var persistence behavior; this change intentionally preserves existing var handling and does not address that issue.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly names the HIR fix and the uninitialized let/const loop-reset bug tied to #6871.
Description check	✅ Passed	The PR description covers the problem, cause, test coverage, and related issue, even though it uses custom headings instead of the template.
Linked Issues check	✅ Passed	The code and regression test address #6871 by materializing implicit undefined for init-less lexical declarations in loop bodies and preserving var behavior.
Out of Scope Changes check	✅ Passed	The diff stays focused on the loop-reset fix, its regression test, and a related changelog note without unrelated feature work.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/6871-uninit-let-loop-reset

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

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