I should provide a little more info I suppose. The Duration class that makes 1.year.ago and friends accurate broke the tests for DateHelper, since the tests depended on the values generated by 1.year etc to be wrong. The patch adjusts the ranges of the minutes that map to which strings, rather than changing the tests to use to_f on the results of 1.year etc. If the latter approach is preferred, I'll whip up another patch for it.
Applied. Thanks for staying on top of that.