pg_rollup

TODO

easy:

1. sorting with a btree index only scan

2. use count(*) to implement aggregateless tables

medium:

1. better parsing https://github.com/pganalyze/queryparser/tree/master/extension

   1. filter syntax on aggregates

2. multiparameter aggregate functions (will affect tdigest, datasketches, allows implementing lots of built-in aggs)

3. use aggregate combining functions by default

4. better tracking of dependencies using postgres built-in features

5. benchmarks

6. good interface for settings

   1. allowing deletes for non-groups

hard

1. outer joins

2. subqueries