Thoughts on Gremlin / TinkerPop support?

[tsanton]

Hi LadybugDB team 👋

First off: sorry for broaching the subject, but I guess someone had to surface it! 😅

I wanted to hear your current thinking on Gremlin / Apache TinkerPop support.
I'm somewhat aware of the history here —this came up on Kuzu:

• Kuzu #4172

I believe the gist was that the team wasn't keen on Gremlin as a standard and leaned toward Cypher plus possibly a DuckDB-style programmatic query-builder API. So I'm asking partly to understand whether LadybugDB inherits that stance unchanged, or whether the project is open to reconsidering.

Where we're coming from:

• We run an OLTP graph database for reading and writing transactional data —graphs are core to our product, not an analytics side-project.
• Our stack is built on Gremlin and its SDKs, and the team is proficient with it.
• Critically: we have never written a single line of Cypher.

That last one is not a knock on Cypher — it's just where our skills and existing codebase sit. The query language therefore is a real adoption gate for us.

What I'm actually trying to learn from you:

1. Is Gremlin/TinkerPop categorically off the table for LadybugDB, or is it something you'd entertain given enough community interest (or contribution)?
2. If it's off the table, what's the core reasoning — is it primarily a philosophical preference for declarative querying over imperative traversals, an architectural mismatch with the vectorized/columnar execution model, or simply maintainer bandwidth?
3. Is there a middle path you'd be more open to — e.g., a Gremlin-to-Cypher translation layer, a TinkerPop-compatible adapter that sits on top, or the programmatic query-builder API that was floated for Kuzu?
4. For teams that are Gremlin-native, what would you honestly recommend today?

Again, just to reiterate: I mostly want to understand your reasoning and roadmap. If the answer is "Cypher only, by design," that's a completely fair and useful answer; it just helps us a lot to hear it directly :)

[adsharma]

Still maintain the same posture as Kuzu. Focus on Cypher in the main repo. But...

some things have changed in the last 3 years. We now have significantly higher coding horse power, even if the human attention remains limited.

It's possible to translate method chaining queries in general and gremlin in particular to Cypher and execute them. A proof of concept extension linked.

Welcome contributions in maintaining and enhancing it.

[adsharma]

#564

[tsanton]

Hehe, that really put a grin on my face 😁 Thanks for the PR! Unfortunately I am not in any form proficient at neither C++ nor DuckDB/LadybugDB extensions, so I will be of little help here 😬

Thought from what I am able to read I do think the optimistic read is that we don't need to translate all of Gremlin to generate something very useful!
With some help from AI I find that the navigational, everyday (assumed) OLAP steps seem to map onto Cypher pretty naturally (V(), E(), has, hasLabel, out/in/both, outE/inE, otherV, values, dedup, order().by(), limit, count, where, as/select) :)
With those steps I'd personally be able to write my way around many OLAP-question!

From previous knowledge, although sparse, I am curious about whether the parsing could lean on something TinkerPop already maintains: I believe they ship an official grammar for the language (ANTLR)

There also seems to be a path where we skip parsing text entirely and consume Gremlin bytecode, which might be the simpler route — possibly scoping to targeting Gremlin 4.0 GA?

My thinking on the way forward is that it'd be worth getting eyes from someone deeper in the TinkerPop world early.
I don't know anyone personally, but I do know of a few names within the community that I believe could give some meaningful feedback on my feeble thoughts 😅

Most importantly! What are your thoughts on the way forward @adsharma?

Again, love the initiative on the PoC: it made a DuckDB/SQL- and Gremlin enthusiast very excited 🤗

[adsharma]

Gremlin bytecode

The byte code interpreter is definitely not coming to the core. Ladybug binaries are 15MB vs 62MB for DuckDB. We want to keep it that way.

ANTLR grammar is probably a better idea than the adhoc one used in the PR.

[adsharma]

Also see the typed programming language support discussion:

https://x.com/lbugdb/status/2061106452635554129

They follow syntax similar to gremlin, but more approachable for people writing idiomatic code in their respective programming languages.