[proggeramlug]

Summary

Function.prototype[Symbol.hasInstance] was undefined in Perry — typeof Function.prototype[Symbol.hasInstance] reported "undefined", any reflective use threw, and a class with a custom static [Symbol.hasInstance] was the only way to reach the protocol. This is the missing-method slice of the #3662 "builtins must throw / behave per spec" cluster (the Function.prototype.{apply,call,bind} not-callable checks landed in #4073).

Before

typeof Function.prototype[Symbol.hasInstance]              // "undefined"
Function.prototype[Symbol.hasInstance].call(undefined, {}) // threw TypeError

After (byte-identical to node --experimental-strip-types)

typeof Function.prototype[Symbol.hasInstance]              // "function"
Function.prototype[Symbol.hasInstance].call(undefined, {}) // false
Function.prototype[Symbol.hasInstance].call(Dog, new Dog()) // true

What changed

• global_this.rs — install_function_has_instance_symbol installs the method on Function.prototype, keyed by the real well-known Symbol.hasInstance (via js_object_set_symbol_property, not an @@-string own property that would leak into getOwnPropertyNames).
• instanceof.rs — function_prototype_has_instance_thunk + ordinary_has_instance. OrdinaryHasInstance differs from the instanceof operator: a non-callable this yields false, not a TypeError. It reuses the operator's full constructor-resolution path (js_instanceof_dynamic) with a thread-local flag (SUPPRESS_INSTANCEOF_RHS_THROW) that makes the unresolved-RHS tail return false instead of throwing — for this path only.
• js_instanceof_dynamic now also resolves the Map/Set/RegExp/ArrayBuffer global constructor values (mirroring the synthetic class ids the compile-time operator emits), so reflective hasInstance and a dynamic x instanceof ctorVar work for them.

Testing

New gap test test-files/test_gap_3662_function_hasinstance.ts is byte-identical to Node in both the default auto-optimize and PERRY_NO_AUTO_OPTIMIZE=1 builds. Existing instanceof gap tests still pass; the instanceof operator is unchanged.

Known gap (follow-up)

Array/Object/Date constructor values are not yet identified on the dynamic path (they aren't noop-thunk closures, so identify_global_builtin_constructor returns None), so hi.call(Array, []) and x instanceof ArrayVar stay false. The literal-RHS instanceof operator handles them at compile time; this pre-existing dynamic-path gap is left for a separate change.