fix a performance regression due to tL.preAllocate