Comment 1 for bug 563986

I've left some comments on the merge proposal. In short, it looks like it suffers from some of the problems our old aggregate code did: namely the subset of columns we're dealing with may be less unique than the full set of columns in the result set.