fix(plugin): MCOL-6088 This ensures that `SELECT col AS alias ... GRO…

[mariadb-LeonidFedorov]

Fix for GROUP BY Ambiguous Column Reference Bug

Problem Description

When a GROUP BY clause contains a column name that is ambiguous (i.e., it matches both a table column and a SELECT list alias pointing to a different column), ColumnStore was throwing an error:

ERROR 1178 (42000): The storage engine for the table doesn't support MCS-2016: Non supported item 'cat_id' on the GROUP BY list.

The expected behavior is to handle the ambiguity gracefully, similar to how standard MariaDB would handle it (with a warning if appropriate).

Example Query

CREATE TABLE test2 (
  id INT UNSIGNED NOT NULL,
  GBP DOUBLE UNSIGNED NOT NULL,
  cat_id INT UNSIGNED NOT NULL,
  cat_id_level_0 INT UNSIGNED NOT NULL
) ENGINE=Columnstore;

-- This query has an ambiguous GROUP BY reference
-- 'cat_id' could refer to:
--   1. The table column 'cat_id'
--   2. The SELECT alias 'cat_id' (which is cat_id_level_0 AS cat_id)

SELECT cat_id_level_0 AS entity_id, SUM(GBP) AS spend, cat_id_level_0 AS cat_id
FROM test2
GROUP BY cat_id;

Root Cause

In the processGroupBy() function in ha_mcs_execplan.cpp, when processing a FIELD_ITEM in the GROUP BY clause:

1. buildSimpleColumn() was called to create a column reference
2. If this failed (returned NULL) due to ambiguity or other issues, the code immediately set nonSupportItem, triggering the error
3. The code didn't check if there was a matching alias in the SELECT list that could resolve the reference

Solution

Modified processGroupBy() function in ha_mcs_execplan.cpp to handle ambiguous column references more gracefully:

1. When buildSimpleColumn() succeeds, use the table column and check SELECT list by sameColumn() match
2. If table column found in SELECT list, also check for alias match as a fallback
3. When buildSimpleColumn() returns NULL or non-SimpleColumn, match by alias name in SELECT list

[mariadb-SergeyZefirov]

I agree with changes as the solve exactly what need to be solved and thus can be merged.

I have a question about nested aggregate unwrapping, but I do not see it as super important. If we differ in behavior from the server for single aggregate, we also may differ for nested aggregates.

[mariadb-SergeyZefirov]

Can it be recursive? E.g., SUM(COUNT(group_by_col)).