as-bench

as-bench is a runtime-agnostic, statistically-aware benchmarking framework for AssemblyScript.

You write benchmarks once, then run them on the runtime you actually ship to — Node, a standalone WASI runtime like wasmtime, wasmer, or wazero — and get Criterion-style timing with confidence intervals, outlier classification, and baseline comparisons. Your bench file is compiled together with a statistics engine and executed inside WebAssembly, so what you measure is what runs.

It ships as a CLI (asb / as-bench) and an npm dev dependency. The statistics engine runs entirely inside wasm — the host only supplies a clock, settings, and the reporting channel.

What you get

• Benchmarks in assembly/__benches__/**/*.ts with bench(), suite(), and blackbox().
• Real statistics — adaptive warmup, an auto-sized sampling plan, bootstrap confidence intervals, and Tukey outlier classification, ported from as-tral / Criterion.rs.
• Suite deltas — each bench after the first reports its change versus the suite baseline, with a permutation-test verdict.
• Baselines — save a run, compare against it later, or diff two saved runs in CI.
• Profiling — per-function instruction counts, wall-clock self time, or allocation bytes, all from one profile command.
• Any runtime — external WASI runtimes stream results over a framed binary protocol (WIPC); compare several runtimes in one table.

Start here

1. Getting Started — install, scaffold, and run your first bench.
2. Writing Benchmarks — bench, suite, blackbox, throughput, and charts.
3. Statistics & Output — how the numbers are produced and how to read them.
4. Baselines — save and compare runs.
5. Profiling — find where the time, instructions, or bytes go.
6. Runtimes & Modes — run the same benches across runtimes.
7. CLI and Configuration — once the defaults stop being enough.

Project layout

A typical project looks like this:

assembly/
  __benches__/
    example.ts
.as-bench/
  build/        compiled wasm artifacts
  baselines/    saved baselines (commit these)
  charts/       generated SVG charts
as-bench.config.json

If you also need fast JSON for the data structures you benchmark, see json-as.