fix: SQL interface "off-by-one' indexing error with GROUP BY clauses that use position ordinals

[alexander-beedie]

Closes #14704.

We already had all the machinery in place for handling "GROUP BY" clauses¹ that use ordinal position values, but were defeated by a classic "off by one" error (as SQL is 1-indexed). Fixed it and added the missing test coverage. Will look at adding support for ordinal position values in "ORDER BY" clauses separately.

(Having ordinal position values in both clauses is supported by standard PostgreSQL syntax).

Example

import polars as pl

df = pl.DataFrame({
    "a": ["xx", "yy", "xx", "yy", "xx", "zz"],
    "b": [1, 1, 1, 2, 2, 2],
    "c": [99, 99, 66, 66, 66, 66],
})

with pl.SQLContext(frame=df) as ctx:
    df = ctx.execute(
        """
        SELECT NULL::date as dt, c, SUM(b) AS total_b
        FROM frame
        GROUP BY 2, 1
        """
    ).collect()
    
    # shape: (2, 3)
    # ┌──────┬─────┬─────────┐
    # │ dt   ┆ c   ┆ total_b │
    # │ ---  ┆ --- ┆ ---     │
    # │ date ┆ i64 ┆ i64     │
    # ╞══════╪═════╪═════════╡
    # │ null ┆ 66  ┆ 7       │
    # │ null ┆ 99  ┆ 2       │
    # └──────┴─────┴─────────┘

Footnotes

1. https://www.postgresql.org/docs/current/sql-select.html#SQL-GROUPBY ↩

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 81.11%. Comparing base (835d198) to head (8c9a081).

Additional details and impacted files

@@           Coverage Diff           @@
##             main   #15584   +/-   ##
=======================================
  Coverage   81.11%   81.11%           
=======================================
  Files        1367     1367           
  Lines      174934   174934           
  Branches     2530     2530           
=======================================
+ Hits       141890   141902   +12     
+ Misses      32569    32557   -12     
  Partials      475      475           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[codspeed-hq]

CodSpeed Performance Report

Merging #15584 will not alter performance

_{Comparing alexander-beedie:fix-sql-ordinal-group-by (8c9a081) with main (835d198)}

Summary

✅ 22 untouched benchmarks

[reswqa]

Classic "off by one" error 😆