It looks like the APIs differ with carbon data depending on spark version so we should probably pick on version to support initially.