0.1 is not representable in binary floating point. The old behavior was adding magic that shouldn't be there.