ParentFull threadbadmintonbaseba·O(nlogn) is probably reasonable. Why break up a long function then if you are experiencing O(nlogn) scaling of compile time on function size?View on HN