// CHECK: [[ACCUM_NEW_LR:%.*]] = "tf.Mul"([[ACCUM_NEW]], [[LR]])
    // CHECK: [[VAR:%.*]] = "tf.ReadVariableOp"([[VAR_HANDLE]])
    // CHECK: [[VAR_NEW:%.*]] = "tf.Sub"([[VAR]], [[ACCUM_NEW_LR]])
    // CHECK: "tf.AssignVariableOp"([[VAR_HANDLE]], [[VAR_NEW]])

  %0 = "tfl.var_handle"() : () -> tensor<!tf_type.resource>
  %1 = "tfl.dequantize"(%arg0) : (tensor<1x2x1x3x!quant.uniform<i8:f32, 1.0>>) -> tensor<1x2x1x3xf32>
  "tfl.assign_variable"(%0, %1) : (tensor<!tf_type.resource>, tensor<1x2x1x3xf32>) -> ()
  func.return %arg0 : tensor<1x2x1x3x!quant.uniform<i8:f32, 1.0>>

        -> !tf_res
      %var_handle = "tf.VarHandleOp"() {container = "c", shared_name = "v0", device = "/TPU:1"}
        : () -> !tf_res
      "tf.If"(%arg1, %id0, %var_handle) {
          then_branch = @if_then_and_else,
          else_branch = @if_then_and_else, is_stateless = false}
        : (tensor<i1>, !tf_res,
           !tf_res) -> ()
      "tf.If"(%arg1, %var_handle, %id0) {

// CHECK: tf_executor.graph
// CHECK: %[[VAR_HANDLE:.*]], {{.*}} = tf_executor.island wraps "tf.VarHandleOp"() <{{{.*shared_name = "var_0".*}}}>
// CHECK: %[[READ_VARIABLE:.*]], {{.*}} = tf_executor.island wraps "tf.ReadVariableOp"(%[[VAR_HANDLE]])

// CHECK-DAG: %[[CST:.*]] = "tf.Const"
// CHECK-DAG: %[[VAR_HANDLE:.*]] = "tf.VarHandleOp"
// CHECK: "tf.AssignVariableOp"(%[[VAR_HANDLE]], %[[CST]])

// Test that a new save function that wraps the SaveV2 op is created.
// CHECK: func.func private @tf_quant__save(%[[ARG:.*]]: tensor<!tf_type.string>)
// CHECK: %[[VAR_HANDLE:.*]] = "tf.VarHandleOp"()
// CHECK-SAME: {{.*shared_name = "var_0".*}}

  TFE_TensorHandle* var_handle = nullptr;
  int num_retvals = 1;
  executed = false;
  TFE_Execute(op.get(), &var_handle, &num_retvals, status.get());
  ASSERT_TRUE(TF_GetCode(status.get()) == TF_OK) << TF_Message(status.get());
  ASSERT_TRUE(executed);
  auto handle_cleaner = tensorflow::gtl::MakeCleanup(
      [var_handle]() { TFE_DeleteTensorHandle(var_handle); });

func.func @resource_output_gets_control_node() -> tensor<!tf_type.resource<tensor<1xf32>>> {
  %var = "tfl.var_handle"() {container = "", shared_name = "states"} : () -> tensor<!tf_type.resource<tensor<1xf32>>>
  func.return %var : tensor<!tf_type.resource<tensor<1xf32>>>
}
// CHECK-NEXT: %[[VAR:.*]], %[[CONTROL:.*]] = tfl.control_node controls "tfl.var_handle"
// CHECK-NEXT: return %[[VAR]]


// CHECK-LABEL: @tfl_call_once_gets_control_node

  TFE_ContextSetServerDef(ctx, 0, serialized.data(), serialized.size(), status);
  EXPECT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);

  TFE_TensorHandle* var_handle = TestVariable(ctx, 2.0, dev2_name);
  EXPECT_NE(var_handle, nullptr);
  TFE_ContextAsyncWait(ctx, status);
  EXPECT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);

  const string function_def = VariableAddFunction();

  if (TF_GetCode(status) != TF_OK) return nullptr;
  TFE_TensorHandle* var_handle = nullptr;
  int num_retvals = 1;
  TFE_Execute(op, &var_handle, &num_retvals, status);
  if (TF_GetCode(status) != TF_OK) return nullptr;
  TFE_DeleteOp(op);
  if (TF_GetCode(status) != TF_OK) return nullptr;
  CHECK_EQ(1, num_retvals);
  TF_DeleteStatus(status);
  return var_handle;
}

                                  strlen(ResourceHandle::ANONYMOUS_NAME)));

  AbstractTensorHandle* var_handle = nullptr;
  int num_retvals = 1;
  TF_RETURN_IF_ERROR(varhandle_op->Execute(
      absl::MakeSpan(&var_handle, num_retvals), &num_retvals));
  AbstractTensorHandlePtr owned_var_handle(var_handle);
  if (!tensorflow::isa<ImmediateExecutionTensorHandle>(
          owned_var_handle.get())) {