Reified

Benchmarks

This page shows what the compiled JSON codec costs compared with System.Text.Json, and what full boundary parsing costs compared with the trusted codec. Those two comparisons are the ones that decide which path an endpoint should take.

The suites live in benchmarks/Reified.Schema.Benchmarks/CodecSuites.fs. scripts/run-benchmarks.sh prompts before starting the run so you can stop other processes first.

Setup

The measured run used:

  • .NET SDK 10.0.203
  • .NET runtime 10.0.7
  • F# 10.0
  • BenchmarkDotNet 0.15.8

JSON codec

The codec suites measure Reified.Schema.Json — the JSON codec compiled from a Schema<'model> declaration — on a realistic aggregate (seven primitive fields, one nested record, and two collections) against System.Text.Json on the same model. Both suites live in benchmarks/Reified.Schema.Benchmarks/CodecSuites.fs.

Run them:

dotnet run -c Release --project benchmarks/Reified.Schema.Benchmarks -- --filter "*JsonCodecBenchmarks*" "*BoundaryParseBenchmarks*"

Measured with a BenchmarkDotNet short job on the recorded toolchain:

Method Mean Allocated
System.Text.Json Serialize 1.44 us 1.11 KB
Reified Json.serialize 1.55 us 1.44 KB
Reified Json.deserializeBytes 2.85 us 2.46 KB
Reified Json.deserialize 3.10 us 2.84 KB
System.Text.Json Deserialize 3.11 us 2.01 KB

The codec compiles once per schema and runs with no runtime reflection, so it stays on par with System.Text.Json's reflection-based serializer while remaining AOT- and trimming-safe by construction. deserializeBytes skips the string-to-UTF-8 conversion and is the faster decode entry point when the payload already arrives as bytes.

The boundary suite compares the trusted codec against full boundary parsing — JsonDocument to Data to Schema.parse with complete path-aware diagnostics:

Method Mean Allocated
Reified Json.deserialize (trusted path) 3.15 us 2.84 KB
JsonDocument + Data + Schema.parse (boundary parsing) 19.78 us 27.71 KB

That gap is the price of diagnostics, redisplayable structured data, and constraint checking, and it is why the two paths exist: parse untrusted input where the diagnostics pay for themselves, and use the compiled codec for trusted payloads such as internal services, storage, and queues.

Conclusion

  • for trusted payloads — internal services, storage, queues — the compiled codec is the right path, and costs about what a reflection-based serializer costs while staying AOT- and trimming-safe.
  • for untrusted input, pay for boundary parsing: the diagnostics, redisplayable structured data, and constraint checking are the reason the schema exists.
  • the codec compiles once per schema declaration, so the compilation cost is not on the per-payload path.