It has comments:
  // TODO: It should not be necessary to check evac-in-progress here.
  // We do it for mark-compact, which may have forwarded objects,
  // and objects in cset and gets here via runtime barriers.
  // We can probably fix this as soon as mark-compact has its own
  // marking phase.
  if (obj == fwd && _heap->is_evacuation_in_progress()) {
     Thread* t = Thread::current();
    ShenandoahEvacOOMScope oom_evac_scope(t);
    return _heap->evacuate_object(obj, t);
  }

We have separate mark-compact marking phase since JDK-8255019, but above statement is still *not* true, since it can not handle the scenario when concurrent/degenerated GC upgrades to Full GC.