Group VALUES inside parentheses (issue506)

[adarshsm]

Fixes #506.

The bug

VALUES rows are laid out one per line at the top level, but as soon as the
VALUES list sits inside parentheses (a derived table, a CTE body, a JOIN
source) reindent breaks lines inside each row instead:

>>> print(sqlparse.format("SELECT foo FROM (VALUES ('abc', 'foo'), ('def', 'bar'), ('ghi', 'baz')) as tmp (foo, bar)", reindent=True))
SELECT foo
FROM (
      VALUES ('abc',
              'foo'), ('def',
                       'bar'), ('ghi',
                                'baz')) as tmp (foo, bar)

Still reproduces on current master.

Root cause

group_values() is the only grouping step in group() that doesn't recurse
into sublists — it only scans the statement's own tokens. A nested VALUES
therefore never becomes a sql.Values group:

Parenthesis '(VALUE...'
 |- Punctuation '('
 |- Keyword 'VALUES'
 |- Parenthesis '(1, 2)'
 |- Punctuation ','
 ...

ReindentFilter._process_values() never runs for it, and the inner
IdentifierLists are laid out as ordinary column lists (the
tlist.within(sql.Values) check in _process_identifierlist() is false).

The fix

Decorate group_values with @recurse(), like the other grouping functions.
Nested lists now group exactly like top-level ones, so they get the same
layout (with and without comma_first):

SELECT foo
FROM (
      VALUES ('abc', 'foo'),
             ('def', 'bar'),
             ('ghi', 'baz')) as tmp (foo, bar)

I also compared before/after output for WITH v AS (VALUES ...),
JOIN (VALUES ...) AS v (...), IN (SELECT ... FROM (VALUES ...)) and
MySQL's ON DUPLICATE KEY UPDATE a = VALUES(a) — the first three now match
the top-level layout, the last is unchanged, and str(parse(s)[0]) == s
holds for all of them.

Testing

• test_grouping_identifiers: nested VALUES is a sql.Values group.

• New TestFormatReindent::test_values_in_parenthesis (plain and
  comma_first).

• Both fail on master and pass with the fix.

• Full suite: 507 passed, 2 xfailed, 1 xpassed. ruff check sqlparse/
  clean. benchmarks/bench_grouping.py still reports linear on all vectors.

• ran the tests (pytest)

• all style issues addressed (ruff)

• your changes are covered by tests

• your changes are documented, if needed (CHANGELOG)

🤖 Generated with Claude Code