Output add;
  ASSERT_TRUE(
      CreateOutputWithScope("Add", {a, a}, root.WithOpName("add"), &add).ok());
  EXPECT_EQ(add.node()->name(), "add");
  EXPECT_EQ(add.node()->type_string(), "Add");
}

	subnameFieldType, found := rt.FindFieldType("CustomObject.nested", "subname")
	if !found {
		t.Fatal("got field not found for 'CustomObject.nested.subname', wanted found")
	}
	if subnameFieldType.Type.GetPrimitive() != exprpb.Type_STRING {
		t.Errorf("got field type %v, wanted string", subnameFieldType.Type)
	}
	flagsFieldType, found := rt.FindFieldType("CustomObject.nested", "flags")
	if !found {

  bool MatchAndExplain(
      const Node* node,
      ::testing::MatchResultListener* listener) const override {
    if (op && node->type_string() != *op) {
      if (listener->IsInterested()) {
        *listener << "\nexpected op " << *op << " but found "
                  << node->type_string();
      }
      return false;
    }

    if (assigned_device && node->assigned_device_name() != *assigned_device) {

  }
  if (should_ignore) {
    *out_resource_op_kind = std::nullopt;
    return absl::OkStatus();
  }

  const XlaResourceOpInfo* op_info = GetResourceOpInfoForOp(n.type_string());
  if (op_info) {
    *out_resource_op_kind = op_info->kind();
    return absl::OkStatus();
  }

  // We conservatively assume that functions will both read and write resource

  TF_ASSERT_OK(EncapsulateXlaComputationsPass::Encapsulate(&graph, &flib_def));

  std::unordered_map<string, Node*> index = graph->BuildNodeNameIndex();
  string function = index.at("launch0")->type_string();

  // Tests the outer graph is as expected.
  {
    std::unique_ptr<Graph> outer = MakeOuterGraph(flib_def, function);
    GraphDef expected_def;
    outer->ToGraphDef(&expected_def);

  Status status = xla_kernel_creator.CreateKernel(flr_, callsite, &kernel_);
  ASSERT_TRUE(status.ok()) << status.ToString();

  EXPECT_EQ("XTimesY", kernel_->name());
  EXPECT_EQ("XTimesY", kernel_->type_string());

  EXPECT_EQ(2, kernel_->num_inputs());
  EXPECT_EQ(DT_FLOAT, kernel_->input_type(0));
  EXPECT_EQ(DT_RESOURCE, kernel_->input_type(1));
  EXPECT_EQ(DEVICE_MEMORY, kernel_->input_memory_types()[0]);

//
//  - A T to indicate a successful operation.
template <class T>
using StatusOrOptional = StatusOr<std::optional<T>>;

StatusOrOptional<Tensor> TryToGetTensorFromConstOp(Node* n) {
  if (n->type_string() != "Const") {
    return {std::nullopt};
  }

  const TensorProto* proto = nullptr;
  TF_RETURN_IF_ERROR(GetNodeAttr(n->def(), "value", &proto));
  Tensor tensor(proto->dtype());

    // }
    //
    // then `Identity` no longer produces Const+I*V because the InplaceAdd
    // operation only modifies Const/clone_2 in place.

    if (IsInPlaceOp(n->type_string())) {
      return absl::OkStatus();
    }
    nodes.push_back(n);
  }

  // Iterate over a copy of the nodes to avoid iterating over g->nodes() while
  // creating more nodes.
  for (Node* n : nodes) {

      maybe_cluster->remove_prefix(absl::string_view("cluster_").size());
      TF_RET_CHECK(absl::SimpleAtoi(*maybe_cluster, &cluster));
      clustered_nodes++;
    }
    clusters[cluster][n->type_string()]++;
    cluster_size[cluster]++;
  }

  string result =
      absl::StrCat("Clustered nodes: ", clustered_nodes,
                   "\nUnclustered nodes: ", cluster_size[kNoCluster],

    const Operation& op, const std::vector<Output>& grad_inputs,
    std::vector<Output>* grad_outputs) {
  ops::GradFunc grad_fn;
  TF_RETURN_IF_ERROR(registry_->Lookup(op.node()->type_string(), &grad_fn));
  TF_RETURN_IF_ERROR(grad_fn(scope_, op, grad_inputs, grad_outputs));
  TF_RETURN_IF_ERROR(scope_.status());
  return absl::OkStatus();
}