> So, if there are no objections, I'm going to reapply Jean-Luc's > original canonicalization; if other regressions are reported and look > important enough to address, then I can use them as test cases to > implement the cost model portion of my code and submit it for review. > Done in r186533.