test(async): #1013 — pin Promise.all + array destructure regression

[proggeramlug]

Closes #1013.

Summary

The reporter's bug — const [a, b] = await Promise.all([fn1(), fn2()]) returning a === undefined / b === undefined — is already fixed on current main as a side effect of two PRs that landed after the reported HEAD (13dc587a / v0.5.1004):

1. fix(hir): #1001 — restore built-in static identity in member position (regression from #973) #1007 (fix(hir): #1001 — restore built-in static identity in member position) collapses Promise.all from PropertyGet { PropertyGet { GlobalGet(0), "Promise" }, "all" } to PropertyGet { GlobalGet(0), "all" }.
2. fix: address #945 scalar method allocation regression (rebased from #980) #1030 (fix: address #945 scalar method allocation regression) added is_global_constructor_expr in the codegen Promise-static dispatch, which accepts both the shallow (GlobalGet(0)) and deeper (PropertyGet { GlobalGet(0), "Promise" }) receiver shapes.

This PR adds a regression test (test-files/test_issue_1013_promise_all_destructure.ts) that pins the byte-for-byte node match. No code change.

Root cause (pre-fix)

At 13dc587a, the HIR for Promise.all([fetchA(), fetchB()]) was PropertyGet { object: PropertyGet { GlobalGet(0), "Promise" }, property: "all" }. The codegen Promise-static dispatch only matched matches!(object.as_ref(), Expr::GlobalGet(_)), so this deeper shape fell through to the generic js_native_call_method("all", …) block — which called "all" against the global Promise constructor function. No such method on a function value, returned 0.0. The awaited value was then a number, and [a, b] = 0.0 produced a=0, b=undefined.

Bisect

Commit	Behavior
13dc587a (#1010, reporter HEAD)	a: 0 typeof: number / b: undefined
42ec277c (#949 test harden, parent of #1007)	Still buggy
f78361f1 (#1007 HIR collapse)	Fixed — a: "hello-from-A" / b: {"plan":"pro"}
3856caad (current main)	Fixed

Repro

async function fetchA(): Promise<string> { return "hello-from-A"; }
async function fetchB(): Promise<{ plan: string }> { return { plan: "pro" }; }
async function main() {
    const [a, b] = await Promise.all([fetchA(), fetchB()]);
    console.log("a:", JSON.stringify(a), "typeof:", typeof a);
    console.log("b:", JSON.stringify(b), "typeof:", typeof b);
    console.log("b.plan:", b?.plan);
}
main();

Expected (node --experimental-strip-types):

a: "hello-from-A" typeof: string
b: {"plan":"pro"} typeof: object
b.plan: pro

Test plan

• test-files/test_issue_1013_promise_all_destructure.ts matches node byte-for-byte on current main
• Sanity-passed adjacent async tests (compat: async lowering blocks self-fetch (express/fastify/nestjs/hono) #1021/async function with await returns undefined on every call after the first #1029/async fn early-return followed by an unreached await infinite-loops #1047 fixtures)

No version bump or changelog change.

[andrewtdiz]

This looks like the right root cause for #1013: the bad Promise.all receiver shape explains why sequential awaits worked while tuple destructuring from Promise.all failed.

One coverage suggestion before merge: the original issue’s repro shape used cross-module imported async helpers that awaited and then returned property reads, e.g. return account.access_token. The current test pins same-file async functions returning literals/object literals, which covers the dispatch bug but not the exact reported async/module/property-return shape.

A stronger regression would add a small fixture module exporting getAccessToken() / getUserPlan(), each awaiting an inner lookup and returning user.accessToken / user.plan, then destructure:

const [accessToken, userPlan] = await Promise.all([
  getAccessToken("u1"),
  getUserPlan("u1"),
]);

Not a blocker if the intent is only to pin the bisected root cause, but it would guard the full #1013 surface more directly.