@@iterator leaks as a string own key, and a well-known-symbol prototype replacement is ignored

[proggeramlug]

Two defects, neither of which is a named method.

const R = class { *[Symbol.iterator]() { yield 1; yield 2; } };
console.log(Object.getOwnPropertyNames(R.prototype).join(","));
// node : constructor
// perry: constructor,@@iterator      <-- a SYMBOL key leaking as a STRING key

class D { *[Symbol.iterator]() { yield 1; yield 2; } }
(D.prototype as any)[Symbol.iterator] = function* () { yield 70; yield 71; };
console.log([...new D()].join(","));
// node : 70,71
// perry: 1,2                          <-- replacement ignored

An own override on the instance (inst[Symbol.iterator] = ...) does work, so it is the prototype write specifically.

Why this is not #9239

#9239 was closed on 2026-09-04 by 4153caea15 ("preserve fresh prototype replacements"). That commit fixes replacing a named method (C.prototype.m = ...) and its fixture test_gap_9239_fresh_instance_prototype_replacement.ts is exactly that shape. Neither defect above involves a named method, and both reproduce on main@12efed1222.

Entry point

crates/perry-hir/src/lower_decl/helpers.rs::is_special_lowered_well_known and the own-key registration beside it (the "@@iterator" spelling).

Untested hypothesis worth checking in the same pass: that predicate's "iterator" arm requires method.function.is_generator, so a non-generator [Symbol.iterator]() { return ...; } may not be exempted at all and would take the generic computed-member path.

Repro

secret-tests/cases/adversarial/recent/28_iterator_protocol_mutation.ts

Done when

Case 28 matches node.