quackmate/

[giscus[bot]]

quackmate/

Image generated with Google Gemini

https://swingbit.github.io/quackmate/

[wldevries]

This is seriously impressive! Have you considered implementing the Universal Chess Protocol to be able to play it against other engines?

[swingbit]

Hi Wouter! :)
Thanks! Not this time, especially because it really isn't a competitive engine. I did implement UCI for Duchess (which you find in this blog, from a few years back), alongside with WASM. Duchess was far less advanced, but in Rust, so a lot faster anyway.

[kryonix]

Hi! Very cool stuff.

Under the rules of SQL recursive CTEs, the recursive member only has access to the rows produced in the immediately preceding step.

This is actually not true for DuckDB. You can choose whether you want the semi-naive semantics (as you have described it, see only the rows from the previous iteration), or if you want to see all rows produced so far. To do so, you can just use recurring.<recursive cte>: duckdb/duckdb#20707

[swingbit]

Thank you so much! Great tip, I'll give it a try!

[swingbit]

It works great. I've updated both code and article. It runs slightly slower than with the unrolled minimax CTEs, but that is understandable and a small price for the restored recursive elegance. Thanks!

[kryonix]

That's awesome! Really nice to see these use-cases in the wild using these novel features. Maybe you can push this even one step further with another trick DuckDB has up its sleeve: The USING KEY semantics for recursive CTEs (https://duckdb.org/2025/05/23/using-key). This allows you to use the recurring table of the recursive CTE as a keyed dictionary, instead of an append-only table. It's even possible to specify the aggregate function you want now: duckdb/duckdb#19481. Again, really cool stuff going on here 🙌🏻 And thanks for the shout out.

[swingbit]

Ah! While reading up about recurring, I did wonder whether USING KEY may help!
My impression though is that the only reasonable key would be a board hash. While that would prune the search space by merging transpositions, it would also essentially turn the search tree into a DAG, breaking the minimax correctness (it does rely on strict parent-child relations). Thanks again for your ideas!

[kryonix]

I agree. Finding an appropriate key may be hard here. I will have to dig a bit deeper into the problem to fully grasp it. It's just almost always a big win, if there actually is a way to use USING KEY—sometimes in terms of space savings, sometimes performance, or both.