spread: `[...arr]` on a `class X extends Array` instance throws "value is not iterable"

[proggeramlug]

[...arr] on a class X extends Array instance throws TypeError: value is not iterable where node spreads it normally.

class MyArr extends Array {}
const sub = MyArr.from([1, 2, 3]);
console.log([...sub]);                  // node: [ 1, 2, 3 ]   perry: TypeError: value is not iterable
console.log(Array.isArray([...sub]));   // node: true          perry: TypeError

Where

array_from_spread_value (crates/perry-runtime/src/array/iterator.rs) has a branch for exactly this:

if crate::array::is_array_subclass_instance(value())
    && !crate::array::array_subclass_has_iterator_override(value())
{
    let snap = crate::array::array_subclass_dense_snapshot(value());
    return crate::value::js_nanbox_get_pointer(snap) as *mut ArrayHeader;
}

so one of is_array_subclass_instance or array_subclass_has_iterator_override is answering wrong for an instance produced by the inherited Array.from (MyArr.from(...)), and the value then falls all the way through to throw_not_iterable. Worth checking whether a directly-constructed new MyArr(...) behaves differently from an inherited-static-produced one — that split is the known-weak "native base-class subclassing" family (CLAUDE.md).

Provenance

Found while building a 32-case [...arr] semantics matrix for #7533. Pre-existing: reproduces byte-identically at f06270d06 (before the #7495/#7516/#7527 rooting stack) and is unchanged by #7540's dense fast path, which deliberately excludes subclass instances — they are GC_TYPE_OBJECT, not GC_TYPE_ARRAY, so they never reach the copy.

Every other case in that matrix matches node, including holes, sparse arrays, past-capacity indices, frozen arrays, index accessors, own @@iterator, strings, Set, Map, generators, arr.values()/arr.entries(), multi-element and call spreads.