Ok so something I figured might never occur -- number of MapperFeature fields filling int -- seems to be happening. Although with 3.0 we can remove/refactor some of them, this will not help 2.x, and we probably still need to add couple more in 2.x timeframe. Because of this, it probably makes sense to upgrade internal bitfield to use long: it should be mostly fine (I think there is just one theoretical gnarly part), and should go in 2.13.