feat: support Node 24

[not-matthias]

What

Adds Node 24 to CI and moves the repo (and therefore the CodSpeed runs) onto it.

• .nvmrc and engines.node go to 24.19.0, which switches the CodSpeed workflow (instrumented, simulation, memory, walltime, walltime-macos) and the CI check/release jobs to Node 24
• the example matrix becomes ["22", "24"], dropping the EOL 18 and 20 entries
• @types/node moves to ^24

Issues found and fixed

Symptom on Node 24	Cause	Fix
every rollup -c rollup.config.ts build fails with SyntaxError: Unexpected identifier 'assert'	Rollup defaults output.importAttributesKey to the legacy assert key when bundling the TS config, and Node 24 removed support for it	--configImportAttributesKey with in the five build scripts, plus the one config still written with assert
all ESM TypeScript benches fail with Cannot use import statement outside a module	esbuild-register's ESM loader no longer intercepts .ts under Node >= 22 module resolution	node --import tsx <file> in packages/tinybench-plugin and the four ESM examples; esbuild-register gives way to tsx there
pnpm --filter @codspeed/tinybench-plugin bench fails with ERR_MODULE_NOT_FOUND .../index.ts	benches/sample.ts imported "..", and tsx resolves a relative directory import to index.ts instead of the package exports	import the plugin by its own name, which still resolves to dist/index.es5.js

CJS examples and @codspeed/benchmark.js-plugin keep node -r esbuild-register; that path works unchanged on 18, 20 and 24.

Why --import tsx and not the tsx CLI

The runner delivers V8 flags with a PATH shim that re-execs $REAL_NODE_PATH $V8_FLAGS "$@", and checkV8Flags reads them back from process.execArgv. The tsx CLI spawns a child through process.execPath carrying only its own loader flags, so that child would bypass the shim and quietly lose --allow-natives-syntax and friends. --import tsx stays in one process.

Verification

On Node 18.20.8, 20.19.2 and 24.19.0:

• pnpm turbo run build --force: green on all three
• Node 24: format lint typecheck 24/24 green, including the @types/node bump
• every bench script the example matrix and the CodSpeed workflow run: green on all three
• the runner's introspection shim replayed against node --import tsx benches in default and simulation mode: flags harvested and present in process.execArgv, no [CodSpeed] missing required flags warning; the same run without the shim does warn, which confirms the check is live

Wall-time impact of the loader swap is a one-off +23ms at startup (65ms vs 42ms). Bench durations and per-task latencies are unchanged within the local noise floor (repeating one command three times spans 9%).

Release prerequisite

Publishing now authenticates to npm over GitHub OIDC, so a trusted publisher must exist before the next v* tag. npm's token exchange is scoped per package name, so all five published packages need their own entry (npmjs.com → package → Settings → Trusted Publisher → GitHub Actions):

Field	Value
Organization or user	CodSpeedHQ
Repository	codspeed-node
Workflow filename	release.yml
Environment name	(leave empty)
Allowed actions	npm publish

Packages: @codspeed/core, @codspeed/benchmark.js-plugin, @codspeed/tinybench-plugin, @codspeed/vitest-plugin, @codspeed/playwright-plugin.

This needs an npm org owner. Until it lands, a tag push fails at the publish step; re-adding NODE_AUTH_TOKEN: ${{ secrets.NPM_TOKEN }} to that step's env restores token publishing.

Note

Unit tests in vitest-plugin and tinybench-plugin fail when the repo is checked out as a git worktree, because getGitDir() looks for a .git directory while a worktree has a .git file, so benchmark URIs pick up an extra path segment. Pre-existing and unrelated to this branch (it reproduces on Node 22 before these commits) and CI is unaffected, so it is left alone here.

[not-matthias]

@greptileai review

[greptile-apps]

Greptile Summary

The PR moves repository automation and compatibility coverage to Node 24 while retaining Node 22 example coverage.

• Migrates ESM TypeScript benchmark entry points from esbuild-register to tsx.
• Converts Rollup configuration files to native ESM and updates JSON import attributes.
• Updates Node/V8 runtime flags, native string conversion, CI actions, and npm publication authentication.

Confidence Score: 5/5

The PR appears safe to merge because no blocking failure remains within the follow-up review scope.

No blocking failure remains.

Important Files Changed

Filename	Overview
packages/core/src/introspection.ts	Extends analysis-mode V8 flags for current Node/V8 garbage collection and Maglev behavior.
packages/core/src/native_core/linux_perf/utils.h	Replaces removed V8 string APIs with v8::String::Utf8Value for Node 24 compatibility.
.github/workflows/ci.yml	Updates setup actions and moves example compatibility coverage to Node 22 and 24.
.github/workflows/release.yml	Updates the release toolchain and switches npm publication from a repository token to trusted-publisher OIDC.
packages/tinybench-plugin/package.json	Uses the in-process tsx loader for the plugin’s ESM TypeScript benchmark.
rollup.options.mjs	Migrates shared Rollup configuration to native ESM for Node 24-compatible config loading.

_{Reviews (2): Last reviewed commit: "chore: clear CI warnings" | Re-trigger Greptile}

[codspeed-hq]

Merging this PR will regress 32 benchmarks

⚠️ Unknown Walltime execution environment detected

Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data.

For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.

⚠️ Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

⚡ 86 improved benchmarks
❌ 32 regressed benchmarks
✅ 120 untouched benchmarks

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	Memory	switch 2	1 B	13,664 B	-99.99%
❌	Memory	test_iterative_fibo_100	400 B	13,736 B	-97.09%
❌	Simulation	switch 1	247.4 µs	1,318.5 µs	-81.23%
❌	Simulation	recursive fibo 10	279.7 µs	1,373 µs	-79.63%
❌	WallTime	test_recursive_cached_fibo_10	1.1 µs	2.5 µs	-57.28%
❌	Simulation	short body 2	122.9 µs	215.9 µs	-43.09%
❌	Simulation	recursive fibo 15 with hooks	381 µs	661.5 µs	-42.41%
❌	Simulation	recursive fibo 15 with hooks	380.9 µs	661.3 µs	-42.4%
❌	Memory	test_iterative_fibo_10	400 B	656 B	-39.02%
❌	Memory	test_recursive_cached_fibo_20	400 B	656 B	-39.02%
❌	Memory	test_recursive_fibo_20	400 B	656 B	-39.02%
❌	Memory	test_recursive_cached_fibo_20	400 B	656 B	-39.02%
❌	Memory	test_recursive_cached_fibo_30	400 B	656 B	-39.02%
❌	Memory	test_recursive_fibo_20	400 B	656 B	-39.02%
❌	Memory	test_iterative_fibo_100	400 B	656 B	-39.02%
❌	Memory	test sync baz 100	400 B	656 B	-39.02%
❌	Memory	test_recursive_cached_fibo_30	400 B	656 B	-39.02%
❌	Memory	switch 2	400 B	656 B	-39.02%
❌	Memory	test sync baz 100	400 B	656 B	-39.02%
❌	Simulation	test_recursive_cached_fibo_10	52.2 µs	83.4 µs	-37.47%
...	...	...	...	...	...

ℹ️ Only the first 20 benchmarks are displayed. Go to the app to view all benchmarks.

Tip

Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing chore/node-24 (534b22a) with main (1999ec9)}

[not-matthias]

@greptileai review