GROUP BY is two things, partitioning and aggregation to a single row within each partition. The relational model doesn't really have a model for partitioning on

GROUP BY is two things, partitioning and aggregation to a single row within each partition. The relational model doesn't really have a model for partitioning on its own so they're fused, but if they were distinct concepts it might make e.g. analytic window functions more composable and less awkward.