I’ve just found this library, says it’s 1.5x faster than nippy and zero dependencies. If that’s true it could potentially be useful for datalevin. Not sure how much datalevin uses nippy. https://github.com/igrishaev/deed#benchmarks Also supports gziped streams which could be interesting. https://github.com/igrishaev/deed?tab=readme-ov-file#gzipped-streams
Nippy is used for data that have no type specified, and also for data values that are bigger than 498 bytes. So the use is not really extensive.
We can certainly experiment with it to see how much it helps.
Ahh, so pretty minor. Probably not worth the effort then in terms of overall impact.