Skip to content

README.md

Latest commit

 

History

History
68 lines (48 loc) · 1.95 KB

README.md

File metadata and controls

68 lines (48 loc) · 1.95 KB
Raw

FileFS cross-language benchmark harness

Measures the shared public API surface across language ports and renders a color heatmap matrix into the root README.md.

Quick start

# From repo root — runs every language runner
python3 bench/run_all.py

# Only selected languages
python3 bench/run_all.py --only go,lua,nodejs,rust

# Re-render README/SVG from existing results (no re-run)
python3 bench/render_matrix.py

Results land in:

  • bench/results/latest.json — machine-readable timings (ns_per_op)
  • bench/matrix.png — heatmap image embedded in the root README
  • bench/matrix.svg — vector source of the same heatmap
  • Root README.md section between <!-- BENCH_MATRIX_BEGIN --> markers

Rendering the PNG requires Pillow (pip install pillow). Without it, the harness still writes the SVG and falls back to that image link.

Workload

See workload.json. Each op is timed alone (with a short warmup). Lower ns_per_op is better. Cell colors are relative within each row (green = faster ports for that API, red = slower).

Implemented runners

All 14 language ports are wired:

c, cpp, dotnet, go, java, kotlin, lua, moonbit, nodejs, python, rust, swift, wasm, zig

Notes:

  • Swift bench executable lives in swift/Sources/FileFsBench (SwiftPM cannot depend on sources outside a package root).
  • MoonBit runs on the JS target (moon run --target js) for high-resolution performance.now timing.
  • Wasm is an in-memory port: some ops (mount_umount, seek_tell_rewind) use the closest available API equivalents.

Adding / extending a runner

  1. Emit a single JSON object on stdout:
{
  "language": "go",
  "runtime": "go1.22",
  "ops": {
    "mkfs": {"ns_per_op": 12345, "iters": 40},
    "write_4kib": {"ns_per_op": 99999, "iters": 40}
  }
}

  1. Register a command in bench/run_all.py (RUNNERS map).

  2. Re-run python3 bench/run_all.py --only <lang>.