It has plenty of rough edges. The query language is limited, portability is limited, docs are still thin, and I would not treat it as a general-purpose replacement for simdjson, yyjson, or RapidJSON.

The benchmark numbers also need caveats. On selective queries jsoon can bail out early, while those libraries are doing full parsing and validation. So the large speedups are mostly about workload and architecture, not a claim that this is just a better JSON parser.

Posting because I think the implementation is interesting and I’d rather get criticism now than after spending more time on it. I’d especially like feedback on correctness, the SIMD/CUDA approach, and whether the benchmarks are framed in a fair way.