// Process output names.
  std::vector<string> output_names_vec;
  if (output_names) {
    output_names_vec.reserve(noutputs);
    for (int i = 0; i < noutputs; ++i) {
      output_names_vec.push_back(string(output_names[i]));
    }
  }

  // Process control output names.
  std::vector<string> control_output_names_vec;
  if (control_output_names) {

  const char task0_name[] = "/job:localhost/replica:0/task:0/device:CPU:0";
  const char task1_name[] = "/job:localhost/replica:0/task:1/device:CPU:0";
  const char task2_name[] = "/job:localhost/replica:0/task:2/device:CPU:0";

  // Create one variable per task.
  TFE_TensorHandle* h0 = TestVariable(ctx, 1.0, task1_name);
  TFE_TensorHandle* h1 = TestVariable(ctx, 2.0, task2_name);

  const char* device_name = "/job:localhost/replica:0/task:0/device:CUSTOM:0";
  const char* first_device_name =
      "/job:localhost/replica:0/task:0/device:CPU:0";
  const char* second_device_name =
      "/job:localhost/replica:0/task:0/device:CPU:1";
  std::array<const char*, 2> underlying_devices{first_device_name,
                                                second_device_name};

  device->arrived_flag = arrived_flag;
  device->executed_flag = executed_flag;
  device->device_name = name;
  device->underlying_device = "/job:localhost/replica:0/task:0/device:CPU:0";
  device->strict_scope_placement = strict_scope_placement;
  TFE_RegisterCustomDevice(context, custom_device, name, device, status);
}

TFE_TensorHandle* UnpackTensorHandle(TFE_TensorHandle* logged_tensor_handle,

      plugin_resource, plugin_resource_name, delete_func);
  auto cc_status =
      cc_ctx->resource_manager()->Create<tensorflow::PluginResource>(
          container_name, plugin_resource_name, cc_resource_ptr);
  status->status = cc_status;
}

void TF_LookupOrCreatePluginResource(
    TF_OpKernelContext* ctx, const char* container_name,
    const char* plugin_resource_name, void** result_plugin_resource,

            underlying_devices_[component_index], underlying_devices_[0])) {
      // Device names are from different address spaces, or we can't figure out
      // whether they are, so we'll fully-qualify everything.
      return underlying_devices_;
    }
  }
  std::vector<std::string> local_names;
  local_names.reserve(underlying_devices_.size());
  for (const DeviceNameUtils::ParsedName& parsed_component :

  return TensorHandlePtr(var_value);
}

void Variable::GeneralAssignment(const char* op_name, TFE_Context* context,
                                 TFE_TensorHandle* value, TF_Status* status) {
  std::unique_ptr<TFE_Op, decltype(&TFE_DeleteOp)> op(
      TFE_NewOp(context, op_name, status), &TFE_DeleteOp);
  if (TF_GetCode(status) != TF_OK) return;
  TFE_OpSetAttrType(op.get(), "dtype", type_);

      TF_GetInputPropertiesListSize(graph_properties, node.name().c_str(),
                                    &num_values, status);
      EXPECT_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
      EXPECT_EQ(num_values, 1);

      std::vector<TF_Buffer*> in_props_buf(num_values, TF_NewBuffer());

      TF_GetInputPropertiesList(graph_properties, node.name().c_str(),

  if (!s.ok()) {
    shape_string = "<error computing shape>";
  } else {
    shape_string = shape.DebugString();
  }
  return absl::StrCat("TensorHandle(shape=", shape_string,
                      ", dtype=", DataType_Name(DataType()),
                      ", type=", FullType().DebugString(), ")");
}

Status AbstractTensorHandle::TensorHandleStatus() const {
  // Tensor handles in current runtime don't carry error info and this method

      tf_gcs_filesystem::Cleanup(filesystem_);
    delete filesystem_;
  }

  std::string GetURIForPath(absl::string_view path) {
    const std::string translated_name =
        tensorflow::io::JoinPath(root_dir_, path);
    return translated_name;
  }

  std::unique_ptr<TF_WritableFile, void (*)(TF_WritableFile* file)>
  GetWriter() {
    std::unique_ptr<TF_WritableFile, void (*)(TF_WritableFile * file)> writer(