"SetOpName called on already built op.");
    }
    if (op_type_.empty()) {
      return errors::FailedPrecondition(
          "GraphOperation::Reset must be called before calling SetOpName.");
    }
    // TODO(b/145674566): We use Graph::NewName to get a unique name here but
    // this may not be consistent with python's naming policy.
    mutex_lock l(g_->mu);

            *   Uses of `tensors()` can be replaced by calling the existing
                methods `tensors_size()` and `tensor(int)`.
            *   Uses of the non-const overload of `nodes_and_registration` can
                be replaced by calling the existing methods `nodes_size()` and
                `context()`, and then calling the `GetNodeAndRegistration`

  // Tell the forward accumulator to watch tensor_id, with a Tensor tangent
  // vector `tangent` of matching shape and dtype. Tangents are the "vector" in
  // "Jacobian-vector product"; `Watch`ing a new Tensor and immediately calling
  // FetchJVP for it would return `tangent`.
  void Watch(int64_t tensor_id, Gradient* tangent);

  // Removes the gradient associated with tensor_id. Should be called when the

  memcpy(ret, debug_str.c_str(), *len + 1);
  return ret;
}

// On success, returns a set of TF_Function instances from `text_proto` of
// GraphDef type. These functions must be deleted by calling TF_DeleteFunction.
//
// If `mutate_proto_func` is non-NULL, run it over each FunctionDef proto,
// before creating a TF_Function out of the possibly mutated proto.