@eddelbuettel Many thanks for putting this out, but why oh why isn't this kind of optimization in R core code?