Groupby and Aggregates - Instructions and Codegen overhaul

Samyak2 commented 1 year ago

Description

We took a second look at the GroupBy instruction and found it very lacking (see #16). This PR represents an overhaul of it.

Partially fixes #24

[x] Add a separate Aggregate instruction to calculate aggregations on a grouped table
[x] Common variables used in codegen are moved to a CodegenContext struct to make passing them around easier.
[x] GroupBy will now produce a grouped table which will be lazily evaluated. Note: the evaluation is not in scope for this PR.
[x] Add tests for these cases.
[x] Update existing tests.
[ ] Non-grouped aggregate functions.

This design was chosen because it introduces minimal changes to the existing architecture. Mainly that only aggregate operations need to be specialized for Grouped tables, the rest remains the same. But with the codegen, the non-aggregate projects are done on the same register indexes. That needs to be fixed.

tyt2y3 commented 1 year ago

Are there test cases illustrating what code will be generated given the sample SQL we use in the blog post?

Samyak2 commented 1 year ago

Not yet. I'm working on it currently.

tyt2y3 commented 1 year ago

Interesting, is there an example for the following too?

select
    col1,
    col3 + 1,
    max(col2 + 1)
from abc
group by col1, col3