zero

Reference

Benchmarks

Benchmark methodology, cases, and metrics.

Benchmarks

The benchmark suite is a regression signal for Zero's goals: small artifacts, fast builds, predictable output, and low overhead.

Run it locally:

npm run bench

The report is written to:

.zero/bench/latest.json

Trend artifacts are written to:

.zero/bench/trends/latest.json
.zero/bench/trends/summary.md

Cases

The current cases include:

  • hello: minimal output
  • add: arithmetic and a helper function
  • structs: shape construction and direct field access
  • params: multiple typed parameters and nested calls
  • buffers: fixed arrays, spans, copy/fill, and byte equality
  • parser: parser-like token metadata and direct predicate checks
  • codec: std.codec loops, varint length, and CRC-32
  • parse: scanner-style digit parsing
  • slices, arena, fallibility, branches: memory views, fixed-buffer allocation style, explicit fallible branching, and branch-heavy loops
  • module-package, rescue: package graph overhead and local rescue lowering
  • fs-resource, mem-copy-fill, zero-hash: focused Zero cases for file-resource metadata, memory helpers, and deterministic byte payload processing

The Zero sources for these cases live under benchmarks/zero. Host targets that do not support a direct executable runner for a case report it as skipped with a reason instead of failing the whole benchmark run.

Metrics

  • buildMs: wall-clock time for the compiler command
  • runMs: median wall-clock time across repeated executable runs
  • runMinMs: minimum wall-clock time across repeated executable runs
  • runRuns: raw per-run wall-clock timings
  • artifactBytes: output executable size
  • compressedArtifactBytes: gzip-compressed executable size
  • peakRssBytes: maximum resident set size when available
  • expectedStdout and outputMatches: semantic checks that catch drift

Claims And Limits

Use these numbers narrowly:

  • build time is represented by buildMs
  • artifact size is represented by artifactBytes
  • compressed size is represented by compressedArtifactBytes
  • run time is represented by the per-case run metrics

Do not treat one local run as a definitive performance claim. Process startup, host load, and cache state all affect results.

Options

Change the repeat count:

ZERO_BENCH_RUNS=<n> npm run bench

Run the sandbox backend explicitly:

ZERO_BENCH_MODE=sandbox npm run bench

Sandbox mode prepares an isolated environment and is useful for CI. Local mode is the default public workflow.