fix: parse EXCLUDE column_list after non-star columns in Redshift

[veeceey]

Summary

Fixes #6963

In Redshift, the EXCLUDE clause can appear at the end of the entire SELECT projection list, not just immediately after *. For example, this is valid Redshift SQL:

SELECT *, 4 AS col4 EXCLUDE (col2, col3) FROM (SELECT 1 AS col1, 2 AS col2, 3 AS col3);

Previously this raised ParseError: Invalid expression / Unexpected token because the parser only recognized EXCLUDE as part of the Star expression (i.e., * EXCLUDE (...)).

Changes

• expressions.py: Added exclude field to Select to store the trailing EXCLUDE column list
• parser.py: Changed _parse_projections return type to include an optional exclude list
• redshift.py (Parser): Override _parse_projections to detect and parse a trailing EXCLUDE clause after all projections
• redshift.py (Generator): Set STAR_EXCEPT = "EXCLUDE" and STAR_EXCLUDE_REQUIRES_DERIVED_TABLE = False for correct Redshift output
• generator.py: Added STAR_EXCLUDE_REQUIRES_DERIVED_TABLE flag; when True (default), transpiles to a derived table with * EXCEPT (...); when False (Redshift), emits EXCLUDE directly
• tsql.py: Updated _parse_projections override to match new return type
• test_redshift.py: Added test cases for identity, normalization, and cross-dialect transpilation

Transpilation behavior

Input (Redshift)	Output dialect	Result
SELECT *, 4 AS col4 EXCLUDE (col2, col3) FROM t	redshift	SELECT *, 4 AS col4 EXCLUDE (col2, col3) FROM t
SELECT *, 4 AS col4 EXCLUDE (col2, col3) FROM t	default	SELECT * EXCEPT (col2, col3) FROM (SELECT *, 4 AS col4 FROM t)

Test plan

• All 926 existing tests pass (excluding test_executor and test_optimizer which require duckdb)
• Added identity tests for EXCLUDE after non-star columns
• Added normalization test (unparenthesized EXCLUDE)
• Added cross-dialect transpilation test
• Verified * EXCLUDE (...) (star-only) still works as before

[veeceey]

Manual Testing Results

I've tested the fix for parsing clauses on aliased expressions in Redshift dialect. All tests passed successfully.

Test 1: Issue SQL Parsing

Original SQL from issue:

SELECT *, 4 AS col4 EXCLUDE (col2, col3) FROM (SELECT 1 AS col1, 2 AS col2, 3 AS col3)

✅ Parses successfully - no more parse errors

Test 2: Round-trip Verification

# Parse and regenerate
parsed = sqlglot.parse_one(sql, dialect='redshift')
regenerated = parsed.sql(dialect='redshift')

✅ Preserves syntax correctly:

SELECT *, 4 AS col4 EXCLUDE (col2, col3) FROM (SELECT 1 AS col1, 2 AS col2, 3 AS col3)

Test 3: Cross-Dialect Transpilation

# Spark
SELECT * EXCEPT (col2, col3) FROM (SELECT *, 4 AS col4 FROM (SELECT 1 AS col1, 2 AS col2, 3 AS col3))

# BigQuery
SELECT * EXCEPT (col2, col3) FROM (SELECT *, 4 AS col4 FROM (SELECT 1 AS col1, 2 AS col2, 3 AS col3))

# DuckDB
SELECT * EXCLUDE (col2, col3) FROM (SELECT *, 4 AS col4 FROM (SELECT 1 AS col1, 2 AS col2, 3 AS col3))

✅ Correctly transpiles to dialect-specific syntax (EXCEPT vs EXCLUDE)

Test 4: Edge Cases

All edge cases passed:

1. Single column EXCLUDE:

   SELECT *, 5 AS col5 EXCLUDE (col1) FROM (SELECT 1 AS col1, 2 AS col2)

   ✅ Parses and regenerates correctly

2. Unparenthesized EXCLUDE:

   SELECT *, 5 AS col5 EXCLUDE col1 FROM (SELECT 1 AS col1, 2 AS col2)

   ✅ Parses and normalizes to parenthesized form

3. Multiple columns in EXCLUDE:

   SELECT *, 6 AS col6 EXCLUDE (col1, col2, col3, col4) FROM (...)

   ✅ Handles multiple columns correctly

4. EXCLUDE on non-wildcard column:

   SELECT col1, col2 EXCLUDE (col3) FROM table1

   ✅ Works on regular column expressions

5. EXCLUDE with complex aliases:

   SELECT *, new_col AS alias_name EXCLUDE (col1) FROM (...)

   ✅ Handles aliases correctly

All tests confirm the fix resolves the original issue and handles various edge cases properly.

[VaggelisD]

Hey @veeceey, we really appreciate the contribution but there's currently an open PR working towards the fix so I'll go ahead and close this one.