Group ROLE as an identifier when used as a column name (issue798)

[adarshsm]

Fixes #798.

Problem

ROLE is a non-reserved keyword, so it can be used as a column name, but it is always tokenized as a keyword and group_identifier only wraps Name/String.Symbol tokens. So in SELECT a, role, b the role member stays a bare keyword while a and b become Identifiers:

>>> il = sqlparse.parse("SELECT a, role, b FROM t")[0].tokens[2]
>>> [(type(i).__name__, str(i.ttype)) for i in il.get_identifiers()]
[('Identifier', 'None'), ('Token', 'Token.Keyword'), ('Identifier', 'None')]

Why not just drop ROLE from the keyword list

That makes the column case work and passes the whole suite, but it silently regresses DDL: with ROLE as a plain name, CREATE ROLE myrole groups as an identifier ROLE aliased to myrole (same for SET ROLE). So a lexer-level change is out.

Fix

Add group_identifier_role, which wraps ROLE in an Identifier only when it is a member of an identifier list (flanked by a comma). Outside a list — CREATE ROLE, SET ROLE, DROP ROLE r1, r2, GRANT ... TO ROLE r — there is no adjacent comma, so it keeps its keyword role. This mirrors the existing m_role = ('null', 'role') special-case already in group_identifier_list.

>>> [(type(i).__name__) for i in il.get_identifiers()]
['Identifier', 'Identifier', 'Identifier']

Tests

Two regression tests in tests/test_grouping.py: role becomes an Identifier in a list, and ROLE stays an un-wrapped keyword in CREATE/SET/DROP ROLE. Full suite passes (496), ruff check sqlparse/ clean, CHANGELOG updated.

Scope note

I scoped this to ROLE to match the existing m_role handling. A few other non-reserved keywords (TYPE, SOURCE, LANGUAGE, LOCATION) have the same latent behaviour in a column list — happy to generalize the approach in this PR or a follow-up if you'd prefer.

[adarshsm]

Rebased onto current master (60cdc64) — the branch had gone conflicting. Force-pushed as a1900a3; the fix itself is unchanged.

The only conflict was in CHANGELOG, which has since been reformatted (single backticks instead of double, and prNNN by author attributions). I kept the current format and restyled my entry to match, adding the attribution:

* Group `ROLE` as an identifier when it is used as a column name in an
  identifier list (e.g. `SELECT a, role, b`), so it appears alongside the
  other columns in `get_identifiers()`, while keeping it a keyword elsewhere
  such as `CREATE ROLE` (issue798, pr872 by adarshsm).

pytest tests/ is 508 passed, 2 xfailed, 1 xpassed, and ruff check sqlparse/ is clean on current master.

No rush from my side — just keeping it mergeable. The scope note in the original description still stands: TYPE, SOURCE, LANGUAGE and LOCATION have the same latent issue, and I am happy to generalise the fix if you would prefer that shape.