to the core.
*   As every PR requires several CPU/GPU hours of CI testing, we discourage
    submitting PRs to fix one typo, one warning,etc. We recommend fixing the
    same issue at the file level at least (e.g.: fix all typos in a file, fix
    all compiler warnings in a file, etc.)
*   Tests should follow the
    [testing best practices](https://www.tensorflow.org/community/contribute/tests)

  TFE_TensorHandle* hcpu = TestMatrixTensorHandle(ctx);
  const char* device_type = TFE_TensorHandleDeviceType(hcpu, status.get());
  ASSERT_EQ(TF_OK, TF_GetCode(status.get())) << TF_Message(status.get());
  ASSERT_TRUE(absl::StrContains(device_type, "CPU")) << device_type;
  int device_id = TFE_TensorHandleDeviceID(hcpu, status.get());

  memset(&option, 0, sizeof(option));
  option.name = const_cast<char*>(name.c_str());
  TF_Filesystem_Option_Value option_value;
  memset(&option_value, 0, sizeof(option_value));
  option_value.type_tag = TF_Filesystem_Option_Type_Buffer;
  option_value.num_values = values.size();
  std::vector<TF_Filesystem_Option_Value_Union> option_values(values.size());
  for (size_t i = 0; i < values.size(); i++) {

  ASSERT_EQ(TF_OK, TF_GetCode(status.get())) << TF_Message(status.get());

  // Test that full type information can be accessed.
  auto outs = unwrap(add_outputs);
  auto h = outs->outputs[0];
  ASSERT_NE(h, nullptr);
  ASSERT_EQ(h->FullType().type_id(), TFT_UNSET);
  ASSERT_EQ(unwrap(inputs[0])->FullType().type_id(), TFT_UNSET);

  // Clean up operation and inputs.
  TF_DeleteAbstractOp(add_op);

                                                   TF_Status* status);

// A handle to a tensor on a device.
//
// Like a TF_Tensor, a TFE_TensorHandle refers to a tensor with a value, shape,
// type etc. Unlike a TF_Tensor, a TFE_TensorHandle may refer to such tensors
// placed in the memory of different devices or remote address spaces.
typedef struct TFE_TensorHandle TFE_TensorHandle;

  std::string device_type = parsed_name.type;
  int device_id = 0;
  if (parsed_name.has_id) {
    device_id = parsed_name.id;
  }

  ctx.device_id = device_id;
  if (device_type == "CPU") {
    ctx.device_type = DLDeviceType::kDLCPU;
  } else if (device_type == "GPU") {
#if TENSORFLOW_USE_ROCM
    ctx.device_type = DLDeviceType::kDLROCM;
#else

      gtl::ArraySlice<AbstractTensorHandle*> gradient_tensors) const override;

  // Calls the passed-in backward function.
  // op_type is the op's name provided in RecordOperation.
  Status CallBackwardFunction(
      const string& op_type, GradientFunction* gradient_function,
      const std::vector<int64_t>& unneeded_gradients,
      gtl::ArraySlice<AbstractTensorHandle*> output_gradients,

// off, and an execution operation for named operations. Often execution will
// simply wrap op execution on one or more physical devices.
//
// device_info is an opaque caller-defined type stored with the custom device
// which is passed to the functions referenced in the TFE_CustomDevice struct
// `device` (execute, delete_device, etc.). It can for example contain the
// names of wrapped devices.
//

    }
    if (raw_device_name) {
      device_name_ = raw_device_name;
    }
    op_type_ = op;
    return absl::OkStatus();
  }
  Status SetOpName(const char* const op_name) override {
    if (op_) {
      return errors::FailedPrecondition(
          "SetOpName called on already built op.");
    }
    if (op_type_.empty()) {
      return errors::FailedPrecondition(

      }
      // Type
      if (const int type_size = default_value.list().type_size()) {
        absl::InlinedVector<unsigned int, 4> type_vector;
        type_vector.reserve(type_size);
        for (int i = 0; i < type_size; ++i) {
          type_vector.push_back(default_value.list().type(i));
        }
        TFE_OpSetAttrTypeList(
            op, attr_name,