It's not perfect, but I think we can deal with the rest of the regressions directly on master, so that this process does not stall.
