const char* first_device = "/job:worker/replica:0/task:1/device:CPU:0";
  const char* second_device = "/job:worker/replica:0/task:2/device:CPU:0";
  const char* device_name = "/job:localhost/replica:0/task:0/device:CUSTOM:0";
  std::array<const char*, 2> underlying_devices{first_device, second_device};
  RegisterParallelDevice(context.get(), device_name, underlying_devices,
                         status.get());

  auto iter = name_ranges.find(arg_name);
  if (iter == name_ranges.end()) {
    status->status = InvalidArgument("Output arg '", arg_name, "' not found");
    return -1;
  }
  return iter->second.second - iter->second.first;
}

int TF_OperationNumInputs(TF_Operation* oper) {
  return oper->node.num_inputs();
}

TF_DataType TF_OperationInputType(TF_Input oper_in) {

  GetFactories()[name] = factory;
}

Status SetDefaultTracingEngine(const char* name) {
  auto entry = GetFactories().find(name);
  if (entry != GetFactories().end()) {
    default_factory = GetFactories().find(name)->second;
    return absl::OkStatus();
  }
  string msg = absl::StrCat(
      "No tracing engine factory has been registered with the key '", name,
      "' (available: ");

  std::string first_device = TFE_TensorHandleBackingDeviceName(
      second_components[0].get(), status.get());
  ASSERT_EQ(second_underlying_devices[0], first_device);
  std::string second_device = TFE_TensorHandleBackingDeviceName(
      second_components[1].get(), status.get());
  ASSERT_EQ(second_underlying_devices[1], second_device);

  // Un-pack the first parallel device's tensor too

                         TFE_TensorHandle* second, TF_Status* status) {
  std::unique_ptr<TFE_Op, decltype(&TFE_DeleteOp)> op(
      TFE_NewOp(context, "Mul", status), TFE_DeleteOp);
  if (TF_GetCode(status) != TF_OK) return nullptr;
  TFE_OpAddInput(op.get(), first, status);
  if (TF_GetCode(status) != TF_OK) return nullptr;
  TFE_OpAddInput(op.get(), second, status);
  if (TF_GetCode(status) != TF_OK) return nullptr;

    TF_Status* status);

// Create and modify a variable placed on a parallel device which composes
// `first_device` and `second_device`.
void BasicTestsForTwoDevices(TFE_Context* context, const char* first_device,
                             const char* second_device);

// Implementations of templated functions ******************************

template <std::size_t num_replicas>

// two component devices, running `x = TPUReplicatedInput(inputs=[a, b])` on the
// parallel device creates a parallel tensor `x` with `a` on the first of
// `underlying_devices` and `b` on the second. Running `a_unpacked, b_unpacked =
// TPUReplicatedOutput(input=x, num_replicas=2)` un-packs the parallel tensor
// into its components.
//
// The filled `device` struct and the allocated `device_info` struct may be

TF_Buffer* TF_CoordinationServiceGetKeyValueWithTimeout(
    const char* key, int64_t key_size, int64_t timeout_seconds,
    TF_CoordinationServiceAgent* agent, TF_Status* status) {
  if (timeout_seconds <= 0) {
    status->status = absl::InvalidArgumentError(
        "TF_CoordinationServiceGetKeyValueWithTimeout invoked with invalid "
        "timeout_seconds <= 0.");
    return nullptr;
  }

    input_tensors->emplace_back(node, idx);

    const auto& iter = input_nodes->find(node);
    if (iter == input_nodes->end()) {
      input_nodes->insert({node, {idx}});
    } else {
      auto& indices = iter->second;
      if (std::find(indices.begin(), indices.end(), idx) != indices.end()) {
        return InvalidArgument("TF_Output ", node->name(), ":", idx,
                               " appears more than once in the input list");

TF_CAPI_EXPORT extern uint64_t TF_NowNanos(void);

// Returns the number of microseconds since the Unix epoch.
TF_CAPI_EXPORT extern uint64_t TF_NowMicros(void);

// Returns the number of seconds since the Unix epoch.
TF_CAPI_EXPORT extern uint64_t TF_NowSeconds(void);

// Populates a TF_ThreadOptions struct with system-default values.
TF_CAPI_EXPORT extern void TF_DefaultThreadOptions(TF_ThreadOptions* options);