Skip to content

benchmarks.md

Latest commit

 

History

History
42 lines (31 loc) · 1.34 KB

benchmarks.md

File metadata and controls

42 lines (31 loc) · 1.34 KB

Benchmarks

Use benchmarks to detect performance regressions in random generation, encoding, sampling, distributions, and identifier helpers. Do not make performance claims without recording the environment and command output.

Run all benchmarks with:

make bench

The target runs:

go test -run=^$ -bench=. -benchmem ./...

For release notes, record:

  • Go version from go version
  • OS and architecture
  • CPU model if available
  • command used
  • package-level benchmark output relevant to the change
  • comparison baseline, usually the previous release tag

If a release changes buffering, derived streams, numeric sampling, string/token encoding, or identifier generation, include a short benchmark summary in the release notes even when results are neutral.

Baseline records

Keep benchmark baselines only when they are useful evidence for a performance-sensitive release or follow-up comparison. Name versioned records as docs/benchmarks-vMAJOR.MINOR.PATCH.md, link them from the release notes or changelog entry that needs the evidence, and summarize the result instead of copying full output when a short excerpt is enough.

Retire or merge old baseline files when they no longer support an active comparison. The benchmark command, Go version, OS/architecture, CPU, and relevant package output must stay with any retained baseline.