v7.0.0-beta.0

Note

This is a beta. Install it with @beta, e.g.
pnpm add -D @codspeed/tinybench-plugin@beta.

This release removes a source of measurement noise from your WallTime
benchmarks, so a result no longer depends on the single process it ran in.

Highlights

Walltime benchmarks run in fresh processes

Your tinybench and Vitest 5 benchmarks now run in 3 fresh processes each, and
CodSpeed pools their rounds into one result. JIT tier-up, heap sizing and
allocator state no longer carry over from one benchmark into the next.

No benchmark code changes required. Your CPU Simulation and Memory results are
unaffected, as are benchmark.js and Vitest 3/4.

Set the fork count, highest priority first:

1. CODSPEED_WALLTIME_FORKS=<n>
2. Vitest 5, per benchmark: bench(name, { forks }, fn)
3. withCodSpeed(bench, { forks }) or codspeedPlugin({ forks })

CODSPEED_PROCESS_ISOLATION=false runs all benchmarks in one process, as
before.

Breaking Changes

• WallTime benchmark values may shift when you upgrade
• Each fork adds Node.js startup and re-runs the benchmark file's setup, so
  WallTime jobs take longer

What's Changed

• feat: run walltime benchmarks in forks by @not-matthias in #101

Full Changelog: v6.1.1...v7.0.0-beta.0

v6.1.1

Fixes

• Vitest: top-level benches now get matching flamegraph and result URIs
• Core: the running node executable is excluded from callgrind, keeping
  flamegraphs readable when benchmarks run another binary (Bazel toolchains)
• Core: the V8 RNG and string hashes are seeded in walltime mode, so runs
  with random inputs or hash-heavy structures no longer vary between runs
• Core: analysis workloads compile eagerly with Sparkplug, reducing
  measurement noise; optimizing compilers stay disabled
• Core: bundled axios and form-data bumped past known CVEs

What's Changed

• fix(core): compile analysis workloads eagerly with sparkplug by @not-matthias in #104
• fix(core): seed the V8 RNG and string hashes in walltime mode by @not-matthias in #105
• fix(core): exclude the running node executable from callgrind by @not-matthias in #102
• build(core): bump bundled axios and form-data past known CVEs by @GuillaumeLagrange in #115
• fix(vitest-plugin): prefix walltime URIs of top-level benches with the file path by @not-matthias in #103
• docs(core): flag that the backend mirrors the tested Node.js majors by @fargito in #116

Full Changelog: v6.1.0...v6.1.1

v6.1.0

Important

CPU Simulation results change for Vitest 5 benchmarks. This release fixes a bug in @codspeed/vitest-plugin where Vitest 5 export getter tracking ran inside the measured code, inflating the measured cost of benchmarks that read module exports. The plugin now sets benchmark.suppressExportGetterWarnings to true by default, unless you set it explicitly. Even though this is a minor release, expect CPU Simulation results for Vitest 5 benchmarks to differ from those measured with 6.0.0 and 6.0.1. Benchmarks on Vitest 3 and 4 are unaffected.

What's Changed

• Disable vitest 5 tracker warning when running under codspeed by @GuillaumeLagrange in #99

Full Changelog: v6.0.1...v6.1.0

v6.0.1

CodSpeed for Node.js 6.0.1 loads again on Linux arm64 systems with an older libstdc++, such as Ubuntu 22.04 and Debian 12. In 6.0.0, the native addon failed there with version 'CXXABI_1.3.15' not found (#96).

What's Changed

• ci(release): build the linux prebuilds on ubuntu 22.04 by @GuillaumeLagrange in #97

Full Changelog: v6.0.0...v6.0.1

v6.0.0

CodSpeed for Node.js 6.0.0 adds support for Node.js 24 and 26, Vitest 5 and Vite 8.

Highlights

Node.js 24 and 26

The plugins now ship prebuilt native addons for Node.js 22, 24 and 26, and warn in CI when running on an untested major.

Vitest 5

@codspeed/vitest-plugin supports Vitest 3, 4 and 5 with no configuration change. On Vitest 5, declare benchmarks with the bench test fixture. Each benchmark passed to bench.compare() is reported as its own benchmark in CodSpeed:

import { test } from "vitest";
import { fibonacci } from "./fibonacci";

test("fibo", async ({ bench }) => {
  await bench("fibo 10", () => {
    fibonacci(10);
  }).run();
  await bench("fibo 15", () => {
    fibonacci(15);
  }).run();
});

To compare implementations side by side, pass them to bench.compare() instead. Each benchmark in the group is still reported as its own benchmark in CodSpeed.

Check out the docs for more information!

Breaking changes

• Node.js 22 or later is now required.
• The CPU simulation instrument now disables Maglev on Node.js 24 and later, so hot functions tier up as they do on Node.js 22, and suppresses V8's minor GC task. Expect a one-time change in measured values after upgrading.

What's Changed

• build: pin -std=c11 for instrument-hooks core.c by @not-matthias in #84
• feat: support Node 24 by @not-matthias in #88
• ci(release): publish any prerelease tag under its own dist-tag by @not-matthias in #91
• feat(core): warn when running on an untested Node.js version by @not-matthias in #90
• feat(vitest-plugin): add vite 8 support by @adriencaccia in #74
• fix(vitest-plugin): stop importing the deprecated vitest subpaths by @adriencaccia in #75
• fix(vitest-plugin): call the bench setup and teardown hooks in analysis mode by @adriencaccia in #94
• fix(core): load the native addon on every Node major by @not-matthias in #92
• feat: support Node 26 by @not-matthias in #89
• ci: migrate to github default arm runners by @lvaroqui in #95
• Support vitest 5 by @GuillaumeLagrange in #86

New Contributors

• @lvaroqui made their first contribution in #95

Full Changelog: v5.7.1...v6.0.0

v6.0.0-beta.2

What's Changed

• fix(core): load the native addon on every Node major by @not-matthias in #92

Full Changelog: v6.0.0-beta.1...v6.0.0-beta.2

v6.0.0-beta.1

What's Changed

• feat(vitest-plugin): add vite 8 support by @adriencaccia in #74
• fix(vitest-plugin): stop importing the deprecated vitest subpaths by @adriencaccia in #75
• fix(vitest-plugin): call the bench setup and teardown hooks in analysis mode by @adriencaccia in #94

Full Changelog: v6.0.0-beta.0...v6.0.0-beta.1

v6.0.0-beta.0

Breaking changes:

• Raised the minimum required version to Node.js 22
• Added --no-minor-gc-task on Node >= 20
• Added --no-maglev on Node >= 24

What's Changed

This release adds support for Node.js 24.

• build: pin -std=c11 for instrument-hooks core.c by @not-matthias in #84
• feat: support Node 24 by @not-matthias in #88
• ci(release): publish any prerelease tag under its own dist-tag by @not-matthias in #91
• feat(core): warn when running on an untested Node.js version by @not-matthias in #90

Full Changelog: v5.7.1...v6.0.0-beta.0

v5.7.1

What's Changed

• fix: emit perf maps when not parsing the full inlining logs from V8 by @GuillaumeLagrange in #85

Full Changelog: v5.7.0...v5.7.1

v5.7.0

Highlights

• Dump full inlining information for higher optimization tiers in walltime so we can show more functions on the profiler. To test it out, set CODSPEED_WALLTIME_PROFILER=samply env variable in the codspeed action.
• Added support for tinybench v5 and v6 in @codspeed/tinybench-plugin

What's Changed

• Add necessary flags to dump optimized inlining info and fix walltime marker placement by @GuillaumeLagrange in #83

Full Changelog: v5.6.0...v5.7.0