java.lang.instrument.Instrumentation.getObjectSize for java.lang.Class instance

[Krystal Mok]

Hi Peter,

Which version of JDK are you using? Assuming you're using a JDK6/JDK7
version released to date, the actual implementation for
j.l.i.Instrumentation.getObjectSize() is as follows:

jvmtiError
JvmtiEnv::GetObjectSize(jobject object, jlong* size_ptr) {
  oop mirror = JNIHandles::resolve_external_guard(object);
  NULL_CHECK(mirror, JVMTI_ERROR_INVALID_OBJECT);

  if (mirror->klass() == SystemDictionary::Class_klass()) {
    if (!java_lang_Class::is_primitive(mirror)) {
        mirror = java_lang_Class::as_klassOop(mirror);
        assert(mirror != NULL, "class for non-primitive mirror must exist");
    }
  }

  *size_ptr = mirror->size() * wordSize;
  return JVMTI_ERROR_NONE;
} /* end GetObjectSize */

So there's special handling for j.l.Class instances' size: it's
reporting the size of the klassOop (the VM internal type that holds
the metadata). That's why changing the number of fields in j.l.Class
doesn't affect the result of getObjectSize(clazz).

(BTW, the code is a bit different after NoPermGen, but the idea stays the same.)

Regards,
Kris

On Sat, Dec 15, 2012 at 2:18 AM, Peter Levart <peter.levart at gmail.com> wrote:
> Hi,
>
> Does anybody know what the size returned by
> java.lang.instrument.Instrumentation.getObjectSize when passed a Class
> instance actually means? It is different for various Class instances
> (contains static fields of various classes?). It's also quite big (520 bytes
> for Object.class on 64bit JVM without compressed OOPS). If I remove some
> instance fields from java.lang.Class the reported size stays the same.
>
> Regards, Peter
>