// CHECK-SAME: initializers = [@init_func_restore_op]

  func.func @init_func_restore_op() -> () attributes {
      tf_saved_model.initializer_type = "restore_op",
      tf_saved_model.exported_names = ["tf_saved_model.session_initializer_restore_op"]} {
    %cst_0 = "tf.Const"() {value = dense<1.000000e+00> : tensor<2xf32>} : () -> tensor<2xf32>

    %pred = arith.constant false
    cf.cond_br %pred, ^bb1(%v : tensor<!tf_type.resource<tensor<?xf32>>>), ^bb1(%v1 : tensor<!tf_type.resource<tensor<?xf32>>>)
  ^bb1(%either: tensor<!tf_type.resource<tensor<?xf32>>>):

  for (auto f :
       llvm::make_early_inc_range(getOperation().getOps<func::FuncOp>())) {
    auto exported_names = GetExportedNames(f);
    if (exported_names.empty() || !f.isPublic()) continue;

    if (exported_names.size() == 1 && exported_names[0] == f.getName()) {
      // Functions that already export themselves with their MLIR name
      // we can leave alone. This saves one level of indirection.

    absl::Span<std::string> exported_names, MLIRImportOptions options = {});

// Given a V1 SavedModel, returns a MLIR module containing the functions,
// expressed with tf_executor dialect.
absl::StatusOr<mlir::OwningOpRef<mlir::ModuleOp>> ConvertSavedModelV1ToMlir(
    const SavedModelBundle& saved_model, absl::Span<std::string> exported_names,
    mlir::MLIRContext* context, MLIRImportOptions options = {});

  # CHECK-SAME:   %arg1: tensor<2xf32> {tf._user_specified_name = "y", tf_saved_model.index_path = [1]})
  # CHECK-SAME: attributes {{.*}} tf_saved_model.exported_names = ["f0000_function_arity"]
  @tf.function(input_signature=[
      tf.TensorSpec([1], tf.float32),
      tf.TensorSpec([2], tf.float32)
  ])
  def f0000_function_arity(self, x, y):
    return

module attributes {tf_saved_model.semantics} {
  "tf_saved_model.session_initializer"() {initializers = [@init_all_tables]} : () -> ()

  func.func @init_all_tables()
  attributes {tf_saved_model.exported_names = ["__tf_saved_model_session_initializer"]} {
    %cst = arith.constant dense<[1, 2, 3, 4]> : tensor<4xi64>
    %cst_0 = arith.constant dense<["a", "b", "c", "d"]> : tensor<4x!tf_type.string>

    # CHECK: "tf_saved_model.global_tensor"() <{is_mutable, {{.*}} type = tensor<?xf32>, value = dense<[0.000000e+00, 1.000000e+00]> : tensor<2xf32>}> {tf_saved_model.exported_names = ["v1"]} : () -> ()
    self.v0 = tf.Variable([0.], shape=tf.TensorShape(None))
    self.v1 = tf.Variable([0., 1.], shape=[None])

  func.func @f(%arg0: tensor<!tf_type.resource<tensor<?xf32>>> {tf_saved_model.bound_input = @v})
  -> (tensor<?xf32> {tf_saved_model.index_path = []})
  attributes {tf_saved_model.exported_names = ["f"]} {
    %0 = "tf.ReadVariableOp"(%arg0) : (tensor<!tf_type.resource<tensor<?xf32>>>) -> tensor<?xf32>
    return %0 : tensor<?xf32>
  }
}

// -----

// CHECK-LABEL: module

  "tf_saved_model.session_initializer"() { initializers = [@init] } : () -> ()
  // expected-error@+1 {{expects exactly one block in the MLIR function}}
  func.func @init() attributes {tf_saved_model.exported_names = ["__tf_saved_model_session_initializer"], tf_saved_model.initializer_type = "restore_op"} {
    cf.br ^bb1
    ^bb1:
    func.return
  }
}

// -----

module attributes {tf_saved_model.semantics} {

      `tf.Module` to be converted and printed.
    exported_names: A set of exported names for the MLIR converter (default is
      "export all").
    show_debug_info: If true, shows debug locations in the resulting MLIR.
  """
  if exported_names is None:
    exported_names = []

  # Make LOG(ERROR) in C++ code show up on the console.