The suggestion you cite has relevance only as long as deduplication is
the chosen implementation scheme, however, Phillip is arguing that the
deduplication and key reordering logic should be dropped, hence, the
cited reference isn't relevant in light of Phillip's suggestion.
Ok!
However, Phillip has presented a good reason to prefer
"output in order requested, no deduplication", and I do find his
argument compelling.
I'll do that! Thanks!