ZioJson is 3x slower than jsoniter using a specific benchmark

[mbaechler]

Mateusz Kubuszok gave a talk about auto derivation at scala.io 2024 and shared the results here https://github.com/MateuszKubuszok/derivation-benchmarks

I proposed a PR to add ziojson here MateuszKubuszok/derivation-benchmarks#3

Results are decent but there's a 3x difference with jsoniter.

I seems like there's room for improvement regarding performances.

[plokhotnyuk]

Those benchmarks (I mean the data model and the related JSON sample) are more for testing of overhead related to derivation, not JSON parsing and serialization itself.

More realistic benchmarks you can find in jsoniter-scala-benchmark.

Here are results for both benchmark types: real-world message samples (GeoJSON, Google Maps API, GitHub Actions API, Open RTB, etc.) and synthetic ones to check reading/writing of different data types.

[mbaechler]

Sure, jsoniter-scala-benchmark is more comprehensive.

This one is comparable to NestedStructs.

Fell free to close this issue if you doesn't feel this case is significant.

I was initially interested with the compile time comparison anyway and found nothing useful for now.

[plokhotnyuk]

The derivation-benchmarks is very valuable if you want to focus on derivation overhead.

Using async-profiler and the following options for JMH you can see what is happening under the hood:

-prof "async:dir=target/async-reports;event=cpu;interval=100000;output=flamegraph;libPath=/opt/async-profiler/lib/libasyncProfiler.so" -jvmArgsAppend "-XX:+UnlockDiagnosticVMOptions -XX:+DebugNonSafepoints"