If the production code has to be maximally performant, you'd want the transform. It doesn't copy data around with a bunch of needless assignments. Pipelining the extract/sort/reconstitute in this manner is absolutely the right thing to do in production if production requirements are to keep it fast and with a small footprint. There's no excuse for not commenting it, but strictly-code-speaking there's no reason not to use this for production if the requisite needs are there.