"important to you");
  return EmptyWhileParams();
#else
  if (ninputs == 0) {
    status->status =
        InvalidArgument("TF_NewWhile() must be passed at least one input");
    return EmptyWhileParams();
  }

  TF_Graph* cond_graph = TF_NewGraph();
  TF_Graph* body_graph = TF_NewGraph();
  cond_graph->parent = g;
  cond_graph->parent_inputs = inputs;

  //   Outputs:
  std::vector<TensorHandlePtr> op_outputs_ TF_GUARDED_BY(execution_mutex_);
  // TF_Status is an incomplete type and so can't be stack allocated. To avoid
  // unnecessary allocations each Execute call, we keep one heap-allocated
  // version for the thread.
  StatusPtr status_ TF_GUARDED_BY(execution_mutex_);

  const std::string device_;
  ExecutorPtr executor_ TF_GUARDED_BY(execution_mutex_);

  const char task1_name[] = "/job:localhost/replica:0/task:1/device:CPU:0";
  const char task2_name[] = "/job:localhost/replica:0/task:2/device:CPU:0";

  // Create one variable per task.
  TFE_TensorHandle* h0 = TestVariable(ctx, 1.0, task1_name);
  TFE_TensorHandle* h1 = TestVariable(ctx, 2.0, task2_name);
  TFE_TensorHandle* h2 = TestVariable(ctx, 3.0, task0_name);

    tensorflow::ImmediateExecutionTensorHandle* unwrapped_handle =
        tensorflow::unwrap(handles[i]);
    if (tensorflow::CustomDeviceTensorHandle::classof(unwrapped_handle)) {
      // One of the inputs we're trying to pack is on a custom device. We'll let
      // the first custom device we see handle all of the packing.
      auto* custom_device_handle =

  TF_Graph* graph = TF_NewGraph();

  // Make two scalar constants.
  TF_Operation* one = ScalarConst(1, graph, s, "one");
  ASSERT_EQ(TF_OK, TF_GetCode(s)) << TF_Message(s);

  TF_Operation* two = ScalarConst(2, graph, s, "two");
  ASSERT_EQ(TF_OK, TF_GetCode(s)) << TF_Message(s);

  // Add oper.
  TF_Operation* add = Add(one, two, graph, s, "add");
  ASSERT_EQ(TF_OK, TF_GetCode(s)) << TF_Message(s);