namespace mangling_util {
namespace {

using ::mlir::tfg::mangling_util::PrintShortTextProto;

const char kAttributePrefix[] = "tf.";
const char kDataTypePrefix[] = "tfdtype$";
const char kTensorShapePrefix[] = "tfshape$";
const char kTensorPrefix[] = "tftensor$";

}  // namespace

string MangleAttributeName(absl::string_view str) {
  return absl::StrCat(kAttributePrefix, str);
}

    %outputs_0, %control_0 = tf_executor.island wraps "tf.Placeholder"() {device = "", dtype = "tfdtype$DT_FLOAT", name = "y", shape = "tfshape$dim { }"} : () -> tensor<0xf32>
    %outputs_1, %control_1 = tf_executor.island wraps "tf.TPUReplicatedInput"(%outputs_0) {N = 1 : i64, T = "tfdtype$DT_FLOAT", device = "", name = "input1"} : (tensor<0xf32>) -> tensor<0xf32>

           the TF attribute, and the other for the type and value of the
           attribute value. Attribute value is a string with the format of
           "type$val", where type can be "bool", "string", "tfdtype", "tfshape",
           "tftensor".
           The value serialization format can be found in attr_util.h.

      %out_c, %out_tensor = "tfd.delegate_kernel"(
        %in_c, %in1_tensor, %in2_tensor) {
        _name = "MatMul",

  %cst_0 = arith.constant dense<1> : tensor<3xi32>
  %0 = "tf.Squeeze"(%arg0) {T = f32, _output_shapes = ["tfshape$dim { size: 4 } dim { size: 64 } dim { size: 64 }"], device = "", squeeze_dims = []} : (tensor<4x64x64x1xf32>) -> tensor<4x64x64xf32>

  // CHECK-DAG: [[LINSPACE:%.*]] = chlo.broadcast_add [[MUL]], [[START]] {broadcast_dimensions = array<i64>}
  // CHECK: return [[LINSPACE]]
  %0 = "tf.Const"() {_output_shapes = ["tfshape$"], device = "", dtype = i32, value = dense<4> : tensor<i32>} : () -> tensor<i32>
  %1 = "tf.LinSpace"(%arg0, %arg1, %0) : (tensor<f32>, tensor<f32>, tensor<i32>) -> tensor<4xf32>
  func.return %1 : tensor<4xf32>
}

func.func @testUnimplementedOp() -> (tensor<i32>, tensor<i32>) {
  %0 = arith.constant dense<1> : tensor<i32>
  %1 = arith.constant dense<2> : tensor<i32>
  %2 = "tf.Maximum"(%0, %1) {_output_shapes = ["tfshape$"]} : (tensor<i32>, tensor<i32>) -> tensor<i32>
  %3 = "tf.Minimum"(%0, %1) {random_attr = "hello"} : (tensor<i32>, tensor<i32>) -> tensor<i32>
  func.return %2, %3: tensor<i32>, tensor<i32>

}

// CHECK-LABEL: @erase_tf_var_is_initialized
func.func @erase_tf_var_is_initialized(%arg0 : tensor<!tf_type.resource<tensor<f32>>>) -> tensor<i1> {
  %vh = "tf.VarHandleOp"() {container = "", shape = "tfshape$", shared_name = "x"} : () -> tensor<!tf_type.resource<tensor<f32>>>
  %is = "tf.VarIsInitializedOp"(%vh) : (tensor<!tf_type.resource<tensor<f32>>>) -> tensor<i1>

// Represents a (partially-defined) shape.
typedef struct TF_Shape {
  int num_dims;  // Must be >= -1; -1 represents unknown rank.
  int64_t* dim_sizes;
} TF_Shape;

// Add a new parameter to a TensorFlow Function.
TF_AbstractTensor* TF_AddFunctionParameter(TF_ExecutionContext* func,
                                           TF_DataType dtype, TF_Shape shape,
                                           TF_Status* s);

#include "tensorflow/c/experimental/saved_model/public/signature_def_param_list.h"
#include "tensorflow/c/experimental/saved_model/public/tensor_spec.h"
#include "tensorflow/c/tf_datatype.h"
#include "tensorflow/c/tf_shape.h"
#include "tensorflow/c/tf_status.h"
#include "tensorflow/c/tf_tensor.h"
#include "tensorflow/core/lib/io/path.h"
#include "tensorflow/core/platform/status.h"
#include "tensorflow/core/platform/stringpiece.h"

  for (const std::optional<llvm::ArrayRef<int64_t>>& shape : shapes) {
    TensorShapeProto& tshape = *shape_list.add_shape();
    if (shape.has_value()) {
      for (int64_t dim : *shape) {
        tshape.add_dim()->set_size(mlir::ShapedType::isDynamic(dim) ? -1 : dim);
      }
    } else {
      tshape.set_unknown_rank(true);
    }
  }

  // If shape is already set, override it. This can happen if we import