: (tensor<!tf_type.resource<tensor<f32>>>, tensor<!tf_type.resource<tensor<f32>>>, tensor<f32>)
    -> (tensor<!tf_type.resource<tensor<f32>>>, tensor<!tf_type.resource<tensor<f32>>>, tensor<f32>)
    tf_executor.fetch
  }
  func.return
}

// -----

!tf_res = tensor<!tf_type.resource<tensor<f32>>>
!tf_str = tensor<3x!tf_type.string>

func.func @init_all_tables() {
  %cst = arith.constant dense<"vocab_file_does_not_exist.txt"> : tensor<!tf_type.string>
  %0 = "tf.HashTableV2"() {container = "", device = "", key_dtype = !tf_type.string, shared_name = "hash_table_/tmp/vocab.txt_-2_-1", use_node_name_sharing = false, value_dtype = i64} : () -> tensor<!tf_type.resource>

func.func private @batched_function(%arg0: tensor<1x3xf32>, %arg1: tensor<*x!tf_type.resource>) -> tensor<1x3xf32>
  attributes {tf._input_shapes = [#tf_type.shape<1x3>, #tf_type.shape<*>], tf.signature.is_stateful} {
  // CHECK: [[handle:%.*]] = "tf.VarHandleOp"()
  // CHECK: "tf.ReadVariableOp"([[handle]])
  %0 = "tf.ReadVariableOp"(%arg1) {device = "/device:CPU:0"} : (tensor<*x!tf_type.resource>) -> tensor<1x3xf32>

  } : (tensor<!tf_type.variant>, tensor<!tf_type.resource>) -> tensor<!tf_type.variant>

  func.return %0 : tensor<!tf_type.variant>
// CHECK: "tf.TakeWhileDataset"(
// CHECK-SAME: <{metadata = "", output_shapes = [#tf_type.shape<>], output_types = [!tf_type.string], predicate = @{{.*}}}> {Targuments = [!tf_type.resource, !tf_type.resource, i64, !tf_type.resource, !tf_type.resource, !tf_type.resource, !tf_type.resource, i64]}
}

func.func @init_all_tables() {
  %cst = arith.constant dense<"%FILE_PLACEHOLDER"> : tensor<!tf_type.string>
  %0 = "tf.HashTableV2"() {container = "", device = "", key_dtype = !tf_type.string, shared_name = "hash_table_/tmp/vocab.txt_-2_-1", use_node_name_sharing = false, value_dtype = i64} : () -> tensor<!tf_type.resource>

    %control_23 = tf_executor.island(%control_18) wraps "tf.SaveV2"(%arg1, %outputs_19, %outputs_21, %outputs_17) : (tensor<!tf_type.string>, tensor<1x!tf_type.string>, tensor<1x!tf_type.string>, tensor<20x4096xf32>) -> ()
    %outputs_24, %control_25 = tf_executor.island(%control_23) wraps "tf.Identity"(%arg1) : (tensor<!tf_type.string>) -> tensor<!tf_type.string>

// CHECK-SAME: ([[ARG0:%.+]]: tensor<!tf_type.resource>, [[ARG1:%.+]]: tensor<!tf_type.resource>)
func.func @resource_missing_subtype(%arg0: tensor<!tf_type.resource>, %arg1: tensor<!tf_type.resource>) {
  // CHECK:      "tf.TPUPartitionedInputV2"([[ARG0]], [[ARG1]])
  %0 = "tf.TPUPartitionedInputV2"(%arg0, %arg1) {N = 2 : i64, _XlaSharding = "", partition_dims = []} : (tensor<!tf_type.resource>, tensor<!tf_type.resource>) -> tensor<!tf_type.resource>

#include "tensorflow/compiler/mlir/tensorflow/ir/tf_types.h"

namespace mlir {
namespace TF {

class TensorFlowRegistryEffectInterfaceFallback;

class TensorFlowDialect final : public Dialect {
 public:
  explicit TensorFlowDialect(MLIRContext *context);
  ~TensorFlowDialect() override;

  static StringRef getDialectNamespace() { return "tf"; }

  // Overrides to redirect to tf_type dialect.

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

      %outputs_0, %control_0 = tf_executor.island wraps "tf.VarHandleOp"() {container = "", shared_name = ""} : () -> tensor<!tf_type.resource<tensor<501000x128xf32>>>
      %outputs_1, %control_1 = tf_executor.island wraps "tf.Cast"(%outputs_0) {Truncate = false} : (tensor<!tf_type.resource<tensor<501000x128xf32>>>) -> tensor<*x!tf_type.resource>

      // CHECK: tfrt_fallback_async.batch_function device([[DEVICE:.*]]) @batched_func ([[BATCHED_FUNC_ARG:%.*]])