Repository navigation
(4) class objects: new / instanceof / typeof on a class-object value #1789
Description
Part of #1785. Design: #1772. Depends on #1786.
(4) new / instanceof / typeof on a class-object value
Scope
Make the instance-side operations work when the class is a class-object value (POINTER) rather than INT32(class_id):
new (classObjectValue)(args)→ readclass_idfrom the class object, instantiate via the existing class_id machinery.x instanceof classObjectValue→ resolve the class_id from the value and walk the class chain.typeof classObjectValue === "function"(classes are callable in JS).extends:js_register_class_parent_dynamicalready readsclass_idfrom a POINTER — confirm it stays correct with the class-object representation.
Acceptance criteria
const C = make(x); const c = new C(); c instanceof C === true.typeof make(x) === "function".- effect's
class X extends make(...)instances (where used) behave correctly. - No regression to
new/instanceofon top-level class declarations.
Blocks
(6).
Activity
fix(hir,codegen): class expressions → heap class objects with per-eva…
proggeramlug commentedon May 25, 2026
Implementation blueprint (after #1787/#1788 in PR #1793)
Current state on a class-object VALUE:
typeof C→"object"(need"function"),new C()→ no-op,c instanceof C→ false.Keystone: a per-object class discriminator. A class object and its instances SHARE
class_id(= template), soclass_idalone can't distinguish them —typeofmust return"function"for the class object but"object"for an instance. AddOBJECT_TYPE_CLASS(next toOBJECT_TYPE_REGULAR=1,OBJECT_TYPE_ERROR=2incrates/perry-runtime/src/error.rs) and stamp it on theClassExprFreshallocation (codegenexpr/static_field_meta.rs, afterjs_object_alloc).Blast radius is bounded (~8
object_type != REGULARgates): own-field get/set are gated behindgc_type == GC_TYPE_OBJECT/class_id != 0checks the class object already satisfies, so flippingobject_typeis safe for the foundation's own-field reads (verified while building #1787). Two caching/typed-feedback gates (field_get_set.rs:2093,typed_feedback/guards.rs) should treat CLASS like REGULAR.Then:
typeof:js_value_typeof→ if POINTER object withobject_type == OBJECT_TYPE_CLASS→"function".new (classObjValue)(): readclass_idfrom the POINTER'sObjectHeaderand route through the existing class_id instantiation machinery.instanceof: resolveclass_idfrom the POINTER and walk the class chain (js_register_class_parent_dynamicalready reads class_id from a POINTER — confirm the instanceof path does too).Coordinate with #1790: the same
OBJECT_TYPE_CLASSdiscriminator + the static-field/prototype slots are what the #1096 layout descriptor must mark as pointer-bearing. Recommend landing the discriminator once, shared between this and #1790, rather than twice.fix(runtime,codegen): new / instanceof / typeof on a class-object val…
fix(runtime): construct instances from class-ref values held as first…
proggeramlug commentedon May 27, 2026
Done —
new/instanceof/typeofon a class-object value is shipped and merged in PR #1795, byte-identical to node with zero regressions:OBJECT_TYPE_CLASSdiscriminator (stamped on theClassExprFreshallocation viajs_object_mark_class) so a class object is told apart from its instances, which share the template's class_id.typeof aClassObject === "function";new (classObjectValue)()reads class_id from the POINTER and instantiates (methods dispatch);x instanceof classObjectValueresolves class_id from the POINTER and walks the chain.Gap test:
test_gap_class_expr_new_instanceof.ts. Note: per-evaluation class objects share the template's class_id, so cross-evaluationinstanceof(onemake()result vs a different one) isn't distinguished — inherent to the shared-class_id model. TheOBJECT_TYPE_CLASSdiscriminator is also what #1790's GC layout descriptor should reuse. Closing as complete (part of epic #1785).