TracingContext* tracing_ctx = dyn_cast<TracingContext>(unwrap(func));
  if (!tracing_ctx) {
    tsl::Set_TF_Status_from_Status(
        s, tensorflow::errors::InvalidArgument(
               "TF_AddFunctionParameter must be called on a TracingContext."));
    return nullptr;
  }
  tensorflow::PartialTensorShape partial_shape;
  if (shape.num_dims != -1) {
    DCHECK(shape.dim_sizes != nullptr);

  opts->tensor_id_data.push_back(src_name);
  const string& src_name_str = opts->tensor_id_data.back();
  // We don't need to store dst's name in tensor_id_data, since `dst` must
  // outlive the ImportGraphDef call.
  opts->opts.input_map[TensorId(src_name_str, src_index)] = ToTensorId(dst);
}

void TF_ImportGraphDefOptionsRemapControlDependency(

      return absl::Status(absl::StatusCode::kFailedPrecondition,       \
                          "struct_size field in " #STRUCT_NAME         \
                          " must be set to " #SIZE_VALUE_NAME ".");    \
    }                                                                  \
  } while (0)

#define VALIDATE_MEMBER(STRUCT_NAME, STRUCT_OBJ, NAME)           \

  void* data = dl_tensor->data;

  if (dl_tensor->byte_offset != 0) {
    status->status = tensorflow::errors::InvalidArgument(
        "Unsupported byte_offset (", dl_tensor->byte_offset,
        ") from DLPack, must be zero");
    return nullptr;
  }

  size_t total_bytes = dl_tensor->dtype.bits / 8;
  for (int i = 0; i < num_dims; i++) {
    total_bytes *= dims[i];
  }

  if (dl_tensor->strides != nullptr &&

        // copies/broadcasts. Other implicit copies may be supported eventually,
        // but need special handling for gradients (gradient of copy-on is not
        // just copy-off but includes a sum) and consideration of performance.
        //
        // TODO(allenl): There may be smarter ways to do this copy in some
        // cases, i.e. with a collective broadcast. We'll need to be careful

  tensorflow::Status status = InitGraphPlugin(plugin_init);
  ASSERT_EQ(status.code(), tensorflow::error::FAILED_PRECONDITION);
  ASSERT_EQ(
      status.message(),
      "'device_type' field in TP_OptimizerRegistrationParams must be set.");
}

TEST(Grappler, OptimizeFuncNotSet) {
  auto plugin_init = [](TP_OptimizerRegistrationParams* const params,
                        TF_Status* const status) -> void {
    TF_SetStatus(status, TF_OK, "");

        // artificial restriction and require that when num_opers=-1, such
        // nodes must have a single output.
        if (node->num_outputs() != 1) {
          return InvalidArgument(
              "When `num_opers` is set to -1, nodes referenced in `inputs` "
              "must have a single output. Node ",
              node->name(), " has ", node->num_outputs(),

  // Requests that the worker thread execute the specified operation. Blocks
  // until the previously pending operation (a StartExecute without a Join) has
  // finished, if any.
  //
  // `cancellation_manager` must live until after `Join` finishes and pending
  // `is_async` operations finish. In addition to allowing the caller to cancel
  // the operation, its `StartCancel` method will be called if op execution

  LoggingDevice* dev = reinterpret_cast<LoggingDevice*>(device_info);
  if (dev->strict_scope_placement && *requested_placement == '\0') {
    TF_SetStatus(s, TF_INTERNAL,
                 "Ops must be placed on the device explicitly, or their inputs "
                 "first copied to other devices.");
    return;
  }
  TFE_Context* context = TFE_OpGetContext(original_op, s);
  if (TF_GetCode(s) != TF_OK) return;

  TensorHandlePtr cpu_value(FloatTensorHandle(3., status.get()));
  ASSERT_EQ(TF_GetCode(status.get()), TF_OK) << TF_Message(status.get());

  // Copying on to a parallel device must be explicit.
  TensorHandlePtr failed_copy_on_result(TFE_TensorHandleCopyToDevice(
      cpu_value.get(), context.get(), device_name, status.get()));
  EXPECT_EQ(TF_GetCode(status.get()), TF_UNIMPLEMENTED);