README.md

Evaluator benchmark

Measure normal execution without a debugger. Each workload is parsed and inferred once, warmed for at least 500 ms and 20 evaluations, then measured in five batches. The output reports median milliseconds and allocated bytes per evaluation. Timing assertions are deliberately excluded from automated tests.

make benchmark-evaluation iterations=20 size=10000

For a comparison with a published engine, extract its NuGet package and compile this same runner against its FScript.Language.dll. Build both runners before measuring and run them sequentially, alternating their order across repetitions:

dotnet build benchmarks/FScript.Evaluation.Benchmarks -c Release -o /tmp/eval-current
dotnet build benchmarks/FScript.Evaluation.Benchmarks -c Release \
  -p:LanguageAssembly=/absolute/path/to/published/FScript.Language.dll -o /tmp/eval-baseline

dotnet /tmp/eval-baseline/FScript.Evaluation.Benchmarks.dll 20 10000
dotnet /tmp/eval-current/FScript.Evaluation.Benchmarks.dll 20 10000

Use the same .NET runtime, architecture, and environment for both processes. Avoid other builds or benchmark processes while measuring. DOTNET_TieredCompilation=0 can isolate optimized execution from tiered compilation; report this setting when using it. CLI startup is a separate measurement: time the published and current managed CLI DLLs with the same dotnet executable and input script. Do not compare a native apphost with a managed DLL, or include compilation in the timing.