Star expansion loses position metadata on generated columns

[doripo]

Note: This issue is about optimizer behavior, not dialect-specific parsing/transpilation.

Fully reproducible code snippet (v28.5.0 / main):

import sqlglot
from sqlglot.optimizer.qualify_columns import qualify_columns

sql = 'SELECT * FROM t'
schema = {'t': {'a': 'INT', 'b': 'VARCHAR'}}

parsed = sqlglot.parse_one(sql)
print(f"Star meta: {parsed.expressions[0].meta}")
# {'line': 1, 'col': 8, 'start': 7, 'end': 7}

qualified = qualify_columns(parsed, schema=schema)
for col in qualified.expressions:
    print(f"{col}: meta={col.meta}")
# t.a AS a: meta={}
# t.b AS b: meta={}

Expected

Expanded columns inherit the star's position metadata.

Actual

Expanded columns have empty meta={}.

References

• Expression.update_positions() exists for copying position metadata (used in parser, BigQuery dialect)
• tests/test_parser.py:985 verifies star position metadata is preserved during parsing
• Root cause is in _expand_stars() in qualify_columns.py (line 868), but similar issues exist in the pivot, coalesce (USING), and struct expansion code paths within that function

---

Happy to submit a PR if this is considered a bug worth fixing. Would appreciate feedback on whether to address just the common case (regular SELECT *) or the other code paths as well.

[doripo]

Update after investigation:

The root cause is actually in qualify_outputs() (line 940-947), not just _expand_stars().

qualify_outputs wraps non-Alias selections in Alias but doesn't preserve position metadata from the original expression. This affects star expansion because:

1. _expand_stars creates Columns from the star
2. qualify_outputs wraps them in Alias, losing any position info

A fix in qualify_outputs would be more general - preserving position for any expression that gets aliased, not just star expansions. This changes the scope question: rather than "which code paths in _expand_stars", it's now "should qualify_outputs preserve position metadata when wrapping?"

Happy to submit a PR if this approach makes sense.

[tobymao]

I'm not sure this is really worth adding given the column and position of a star expansion isn't real. The expanded columns are generated by the optimizer and aren't part of the original sql.