compile: Array.prototype.forEach(set.add, set) on an Any destructured-for-of receiver throws 'Set.prototype.add called on incompatible receiver' — pi wall #7

[proggeramlug]

Wall #7 of the pi coding-agent bring-up (tracker #6564; prior: #6593, #6604, #6644, #6649, #6651, #6652 all fixed). With the six fixes applied (current: PR #6656), pi-native --help/--version die during module init with:

TypeError: Method Set.prototype.add called on incompatible receiver
    at pi-bundle.mjs:203827

(line from the refreshed --debug-symbols build dist/pi-native-dbg; the stripped build says at <anonymous>.)

Site: @babel/types/lib/index.js (webpack-wrapped CJS factory), the alias-expansion init loop, pi-bundle.mjs:203827:

for (const { types: e4, set: t4 } of n7.allExpandedTypes) for (const r4 of e4) {
  const e5 = n7.FLIPPED_ALIAS_KEYS[r4];
  e5 ? e5.forEach(t4.add, t4) : t4.add(r4);   // ← line 203827
}

Node: Array.prototype.forEach(callback, thisArg) calls the extracted native t4.add (Set.prototype.add) with this = t4 — works. Perry: the receiver reaching the native method fails the Set brand check.

11-line repro (node: sizes: 3 1; perry: the TypeError above, rc=1):

const FLIPPED: any = { A: ["x", "y"], B: null };
const allExpandedTypes: any[] = [
  { types: ["A", "B"], set: new Set() },
  { types: ["B"], set: new Set() },
];
for (const { types: e4, set: t4 } of allExpandedTypes) {
  for (const r4 of e4) {
    const e5 = FLIPPED[r4];
    e5 ? e5.forEach(t4.add, t4) : t4.add(r4);
  }
}
console.log("sizes:", allExpandedTypes[0].set.size, allExpandedTypes[1].set.size);

Already ruled out (each of these works in isolation, byte-identical to node — the bug needs the combined shape, so it is NOT a plain forEach-thisArg gap):

["a","b"].forEach(s.add, s);                       // typed receiver: OK
const t4: any = new Set(); ["a"].forEach(t4.add, t4); // Any receiver: OK
const add = t4.add; add.call(t4, "q");             // extraction + .call: OK
for (const { set: t4 } of rows) ["a"].forEach(t4.add, t4); // destructured for-of: OK
e5 ? e5.forEach(t4.add, t4) : t4.add("z");         // ternary shape: OK

Suspect subsystem: the interaction of nested for-of + object-pattern loop bindings + the two-armed ternary dispatch on an Any receiver — some path re-materializes/wraps t4.add (or the thisArg) so the native Set method's brand check sees a non-Set receiver. Possibly the same family as the alternating t4.add(r4) direct-call arm specializing the method and poisoning the extracted-callback arm (or vice versa).

Also on the road ahead (NOT this wall, hit while probing): the bundle-shape var GaxiosInterceptorManager = class extends Set {} inside a CJS factory, new exportsObj.GaxiosInterceptorManager() then .add(...) → TypeError: add is not a function (builtin-subclass through a class EXPRESSION + property-read new; family of #6232 class extends Array). gaxios instantiates these per Gaxios instance — it will bite once init gets past @babel/types.

Repro binaries: secret-tests/pi-target/dist/pi-native (stripped) and pi-native-dbg (debug symbols, prints the bundle line), both built from PR #6656's branch; pi-native.pre6652 reproduces wall #6 for contrast. HOME=$(mktemp -d) ./pi-native --version.