// "ret0" = "arg1"
    // "ret1" = "arg0"
    tensorflow::Scope s = tensorflow::Scope::NewRootScope();
    Output arg0 = ops::_Arg(s.WithOpName("arg0"), DT_RESOURCE, 0);
    Output arg1 = ops::_Arg(s.WithOpName("arg1"), DT_RESOURCE, 1);
    Output arg2 = ops::_Arg(s.WithOpName("arg2"), DT_INT32, 2);
    auto ret0 = ops::_Retval(s.WithOpName("ret0"), arg1, 0);
    auto ret1 = ops::_Retval(s.WithOpName("ret1"), arg0, 1);

  // TODO(b/74182462): We implement this by creating a new dummy Graph including
  // _Arg nodes, and let CompileGraph walk it. This could be optimized.
  std::unique_ptr<Graph> graph(new Graph(OpRegistry::Global()));

  // First create the actual node we care about computing.
  TF_ASSIGN_OR_RETURN(Node * main_node, graph->AddNode(node_def));

  // Create dummy _Arg nodes. Link these to `node` and also via a control

}

TEST(CreateSaverDefTest, CreateValidSaverDef) {
  // Needs to have a _Arg node with an attribute "tf_saved_model.index_path" =
  // ["__tf_file_prefix"].
  GraphDef graph_def;
  ASSERT_TRUE(TextFormat::ParseFromString(
      R"pb(node {
             name: "foo",
             op: "_Arg",
             attr {
               key: "tf_saved_model.index_path",

}

// Builds XlaRecvAtHost node, and replaces all _Arg nodes with it.
absl::StatusOr<Node*> ReplaceArgNodesWithRecvAtHostNode(
    Graph* g, const string& oc_cluster_name,
    std::vector<DataType>* recv_at_host_dtypes, Node* key_placeholder) {
  // TODO(b/77601805): use out nodes for source node, instead of traversing all
  // nodes.
  std::vector<Node*> arg_nodes = GatherNodesWithType(*g, "_Arg");

// lifted out of If/While/function node. Attribute value will always be boolean
// value "true".
extern const char kXlaIsLiftedArgAttrName[];

// Attribute indicating that this node is a Placeholder node for an _Arg node
// lifted out of If/While/function node. Attribute value will be a string, which
// is the outside compilation cluster name sending the lifted arg node to host.
extern const char kXlaLiftedArgOutsideCompilationAttrName[];

    int GetResultIndexForEdge(const Edge* edge) const;

    // Creates an _Arg node for the src node of edge, and add its index to
    // args_by_src_, if none exists yet. Also adds its index to args_by_dst_,
    // and adds the edge within the subgraph from the _Arg node to the image of
    // the dst node.
    Status RecordArg(const Edge* edge,

// TODO(phawkins): this code does not infer accurate shapes for cyclic graphs.
// `arg_shapes`: user given map from the `index` to shapes of this
// node, where `index` is the `index` attribute of `_Arg` op or `_index`
// attribute of `Placeholder` op.
Status InferShapes(Graph* graph, const std::map<int, InferredShape>& arg_shapes,
                   const tensorflow::FunctionLibraryDefinition* fnlib_def,

  graph.SetConstructionContext(ConstructionContext::kEagerRuntime);
  tensorflow::set_tf2_execution(true);

  ConfigProto config = ConfigProto();
  Scope root = Scope::NewRootScope().ExitOnError();

  Output a = ops::_Arg(root.WithOpName("A"), DT_RESOURCE, 0);
  std::vector<NodeBuilder::NodeOut> inputs({NodeBuilder::NodeOut(a.node())});

  Node* call;
  NameAttrList f_name_attr;
  f_name_attr.set_name(fd.signature().name());
  TF_ASSERT_OK(

                << context->DebugString(handle);
      }
    }

    int index = -1;
    if (n->type_string() == "_Arg") {
      // NOTE: during runtime, Placeholder ops will be replaced as `_Arg` ops.
      // And Args must have `index` attribute.
      TF_RETURN_IF_ERROR(GetNodeAttr(n->attrs(), "index", &index));
    } else if (n->type_string() == "Placeholder") {

      count: 5
    }
    op_histogram {
      op: "Mul"
      count: 4
    }
  }
  unclustered_op_histogram {
    op: "NoOp"
    count: 2
  }
  unclustered_op_histogram {
    op: "_Arg"
    count: 1
  }
  unclustered_op_histogram {
    op: "_Retval"
    count: 1
  }
}
)",
      &expected_auto_clustering_activity);
  EXPECT_EQ(listener()->auto_clustering_activity().DebugString(),