Compiled Cypher: a grouping key read in an order term's body fails, and one in any(y IN … WHERE y = e) groups by its variable (port of drivine4j#505) #316

Port of liberation-data/drivine4j#505 (fixed in liberation-data/drivine4j#507). The part not in #315.

Two shapes that #315's _aggregates_first doesn't handle yet. Both were checked on Neo4j 2026.10.1.

1. An order term that aggregates isn't rekeyed. On #315's branch (355b720), drivine/cypher/sql/sql_cypher_compiler.py:3605-3616 (sorted_by) replaces an order term only at its top level with replaced(sort.expression, sorted_by), never inside a body. The order terms' beside (3562-3574) keeps only variables in keyed. So a property key read in an order term's body leaves its variable unbound:

UNWIND [{a: 1, b: 1}, {a: 1, b: 2}, {a: 2, b: 1}] AS x RETURN x.a AS a, count(*) AS n ORDER BY size([y IN collect(x.b) WHERE y > x.a])
-- here: "x is not defined"; Neo4j: [2, 1], [1, 2]
UNWIND [{a: 1, b: 1}, {a: 1, b: 2}, {a: 2, b: 1}] AS x RETURN x.a AS a, count(*) AS n ORDER BY size([y IN collect(x.b) | x.a]), a
-- here: "x is not defined"; Neo4j groups by x: [1, 1], [1, 1], [2, 1]

2. Neo4j rewrites any(y IN list WHERE y = e) as e IN list, and none(…) as NOT e IN list, where e reads no y. It also reads y IN [e] as y = e. So such a quantifier over an aggregate reads the key at the top level and groups by the property. all() and single() aren't rewritten. With rekeyed (3517), such a quantifier's WHERE groups by the variable:

UNWIND [{a: 1, b: 1}, {a: 1, b: 2}, {a: 2, b: 1}] AS x RETURN x.a AS a, any(y IN collect(x.b) WHERE y = x.a) AS n ORDER BY a
-- here: [1, false], [1, true], [2, false]; Neo4j: [1, true], [2, false]

The same holds for none(y IN collect(x.b) WHERE x.a = y), any(… WHERE y IN [x.a]) and any(t IN collect(m.title) WHERE t = r.role).

What drivine4j's fix does (liberation-data/drivine4j#507, aggregatesFirst in SqlCypherCompiler.kt):

• Order terms go through rekeyed too. An item's name is still read first, with readingKeys (reading_keys there) in bodies. Where an order term's body groups by a property key's variable, that variable is kept beside the keys, as for an item.
• readAsIn marks an any/none quantifier whose WHERE is y = e, e = y, y IN [e] or e IN [y], where e reads no y. rekeyed reads that WHERE through readingKeys, as it does a list comprehension's WHERE.

Tests to port:

• Pin a grouping key read in a nested body, a body without an aggregate, any(… WHERE y = e) and an order term drivine-cypher-spec#43's steps in "Implicit grouping key": the any/none/all steps and the three order-term steps.
• The any/order-term part of CypherQueryContract's "a key read in a body nested in a list comprehension's WHERE, …" test in Read a grouping key in a nested body, a body over no aggregate, any(y IN … WHERE y = e) and an order term, as Neo4j does drivine4j#507.

Update (from drivine-python PR #318, which syncs spec v0.19.0): besides the ORDER BY bodies and any→IN steps, spec v0.18.0's queries.yaml "Implicit grouping key" steps 29–35 still fail on Postgres and Oracle on main, though #315 was meant to cover the nested-body rekeying: they give a row per variable where Neo4j gives one per key. #318 holds them against this issue. The port should find why: compare with drivine4j #507 (readingKeys/rekeyed, merged) and drivine #498, and fix them here, or split them into their own issue if the cause is different.