NodeDef* const node_def = graph_def.add_node();
          *node_def = node->def();
        }
      }
      *graph_def.mutable_library() = graph.flib_def().ToProto();
      if (flags::Global().more_stack_traces.value()) {
        *graph_def.mutable_debug_info() = graph.BuildDebugInfo();
      }
      session->graph->mu.unlock();
      status->status = session->session->Extend(std::move(graph_def));

  options->stack_size = 0;
  options->guard_size = 0;
  options->numa_node = -1;
}

TF_Thread* TF_StartThread(const TF_ThreadOptions* options,
                          const char* thread_name, void (*work_func)(void*),
                          void* param) {
  ::tensorflow::ThreadOptions cc_options;
  cc_options.stack_size = options->stack_size;
  cc_options.guard_size = options->guard_size;

  tensorflow::StackTracesMap stack_traces;
  for (const Node* n : fn_body->graph.nodes()) {
    stack_traces[n->name()] = n->GetStackTrace();
  }

  TF_Function* tf_function = new TF_Function();
  tf_function->record = new tensorflow::FunctionRecord(
      std::move(fdef), std::move(stack_traces), false);

  return tf_function;
}

  CancellationManager* cancellation_manager_ TF_GUARDED_BY(execution_mutex_);
  //   Outputs:
  std::vector<TensorHandlePtr> op_outputs_ TF_GUARDED_BY(execution_mutex_);
  // TF_Status is an incomplete type and so can't be stack allocated. To avoid
  // unnecessary allocations each Execute call, we keep one heap-allocated
  // version for the thread.
  StatusPtr status_ TF_GUARDED_BY(execution_mutex_);

  const std::string device_;

//
// This method uses a temporary buffer to hold contents.
int CopyFileContents(int dst_fd, int src_fd, off_t size) {
  // Use a copy buffer of 128KB but don't store it on the stack.
  constexpr static size_t kPosixCopyFileBufferSize = 128 * 1024;
  std::unique_ptr<char[]> buffer(new char[kPosixCopyFileBufferSize]);

  off_t offset = 0;
  int bytes_transferred = 0;
  int rc = 1;