On Fri, Jan 2, 2026 at 6:53 PM Henson Choi <assam258@gmail.com> wrote:
>
> When I raise issues about differences from Cypher, even when I sense my suggestions
> might be overreaching, I do so for two reasons: (a) I don't have access to the
> SQL/PGQ standard document, and (b) my implementation-level experience is primarily
> with Cypher. However, if I feel this dissonance, users coming from Cypher will
> likely experience the same confusion. I hope you'll consider this user perspective
> valuable, even when the technical decisions differ.
I think all these issues are worth raising on pgsql-hackers. Many
users who are using Cypher may want to use PostgreSQL + SQL/PGQ as
their graph database. All these inputs are valuable. Let me summarise
my thoughts briefly over here.
IMO, at a higher level, we could leverage Apache/AGE-like extensions
to provide Cypher compatibility on top of SQL/PGQ. The required
functionality from SQL/PGQ (directly or indirectly) can be implemented
in the core. For example, the core will throw an error when a label is
not defined, but Apache/AGE may detect such labels beforehand, avoid
translating the query to SQL/PGQ and return 0 rows in such a case. Or
Apache/AGE may add stable functions for labels() and property_names(),
but use SQL/PGQ catalogs for implementation.
I envision the relation between Apache/AGE and SQL/PGQ evolving
similarly to how the relation between pg_partman and built-in
partitioning evolved. Others may have different opinions. We may even
want to weigh each feature/functionality separately. How to support
Cypher queries on PostgreSQL in the context of SQL/PGQ seems to be a
broader topic worth a discussion of its own.
--
Best Wishes,
Ashutosh Bapat
