Docs Feedback modules/ROOT/pages/appendix/gql-conformance/index.adoc (ref: cypher-25)

[averdeny]

Hi Cypher team,

I'm working on a Python library for parsing and transpiling GQL/Cypher. While building our Neo4j dialect I relied heavily on the conformance documentation. Here are a few things I ran into, plus some features that seem to be missing from the docs.

1. Questions and potential issues

SELECT: Select statements (e.g. SELECT 1) are rejected by Neo4j but I don't see it mentioned on the unsupported mandatory page.

GV39 temporal literals: The type support works, but the GQL literal form DATE '2026-01-01' is rejected (the Cypher constructor date('2026-01-01') works fine). A note on the GV39 entry would help clarify this for implementers.

STDDEV_SAMP / STDDEV_POP: The analogous Cypher page (GF10) spells these STDEV_SAMP / STDEV_POP (single D). The GQL standard uses STDDEV_SAMP / STDDEV_POP (double D). Looks like a typo.

MOD(): GQL defines MOD() for modulus; Cypher uses % instead and rejects MOD(). Not mentioned on any conformance page.

Trailing decimal: RETURN 3. (no fractional digits) is rejected; RETURN 3.0 works. Minor edge case.

DOUBLE vs FLOAT64: The analogous page maps both to Cypher's FLOAT. In the GQL standard FLOAT64 is exactly IEEE 754 binary64 while DOUBLE has implementation-defined precision (>= FLOAT). Since Cypher's FLOAT is 64-bit IEEE 754, the FLOAT64 mapping is precise, but the DOUBLE mapping could use a clarifying note.

2. Possibly undocumented optional features

Through integration testing (parse GQL, generate Cypher, execute against Neo4j 2026 community) I found several optional GQL features that work but don't appear on any of the three conformance pages.

Feature	Description	Test query	Notes
G005	Path search prefix in a path pattern	MATCH SHORTEST 2 (a)-[:KNOWS]->{1,5}(b) RETURN b.name	Prerequisite for G016-G020 which are all listed; likely just missing from the docs
G044	Basic abbreviated edge patterns	MATCH (a)-->(:Person) RETURN a.name	GQL -> / <- / - vs Cypher --> / <-- / --
G111	IS LABELED predicate	MATCH (n) WHERE n:Person RETURN n.name	GQL n IS LABELED Person vs Cypher n:Person
GA05	Cast specification	RETURN toFloat(42)	GQL CAST(x AS FLOAT) vs Cypher toFloat(x) etc.
GB03	Double solidus comments	// comment	Same syntax
GD02	Graph label set changes	MATCH (n) SET n:Intern RETURN n.name	Same syntax
GD04	DELETE with simple expression	MATCH ()-[r:LIKES]->() DELETE r	Same syntax
GE07	Boolean XOR	WHERE n.active = TRUE XOR n.age > 30	Same syntax
GF20	Aggregate functions in sort keys	ORDER BY COUNT(m) DESC	Same syntax
GL01	Hexadecimal literals	RETURN 0xFF	Same syntax
GL02	Octal literals	RETURN 0o77	Same syntax
GL04	Exact number in common notation	RETURN 3.14	Same syntax (trailing-dot 3. fails)
GP02	Inline procedure: implicit scope	CALL { RETURN 42 AS x } RETURN x	Same syntax
GQ12	OFFSET clause	ORDER BY n.name SKIP 3	GQL OFFSET n vs Cypher SKIP n
GQ14	Complex expressions in sort keys	ORDER BY n.age * -1	Same syntax
GQ16	Pre-projection aliases in sort keys	ORDER BY person_name	Same syntax
GQ22	EXISTS predicate: multiple MATCH	WHERE EXISTS { MATCH ... MATCH ... }	Same syntax
GV41	Duration types	RETURN duration('P1Y')	GQL DURATION 'P1Y' literal vs Cypher duration(...) constructor
GV48	Nested record types	RETURN {a: {b: 'hello'}}.a.b	Same syntax

Additionally, G043 (complete full edge patterns) and G045 (complete abbreviated edge patterns) are partially supported. Neo4j handles the directed forms (<-[r]-> for G043, <--> for G045) but rejects the tilde forms (~[r]~, <~[r]~, ~[r]~> for G043 and ~, <~, ~> for G045). Since Cypher only has directed edges, this is expected. Whether these count as supported probably depends on how strictly partial conformance is interpreted.

3. What's working well

Overall the conformance documentation has been genuinely useful for our implementation. The analogous Cypher page in particular: explicit function-name mappings like COLLECT_LIST -> collect and PATH_LENGTH -> length saved us a lot of guesswork, and every mapping we tested works correctly.

Thanks for maintaining these docs.