[not-matthias]

Restart node with the required V8 flags when the launcher bypassed the runner's wrapper

The runner passes the flags from getV8Flags() through a node script on
PATH. Launchers that start node by absolute path never consult it: Bazel's
rules_js runs its toolchain binary through its own shim, so tinybench and
benchmark.js measured without --no-opt, --predictable, the seed flags and
the GC flags. NODE_OPTIONS cannot carry them (node rejects several).

ensureV8Flags() now replaces the process at plugin load, before any benchmark
or instrument-hooks state exists. It appends only the flags missing from
process.execArgv and reuses the current argv, so launcher arguments such as
rules_js's --require .../register.cjs and --preserve-symlinks-main survive.
On Linux it execs /proc/self/exe rather than process.execPath, which
rules_js points at its shell shim; on macOS it uses process.execPath.

The restart runs in a small execv binding in the existing native addon, which
keeps the PID, environment and open descriptors. That matters because the
runner, perf and eBPF attach by PID, and cancellation signals the benchmark
process directly. The binding clears FD_CLOEXEC on stdio first, like node's
own process.execve, otherwise the restarted process writes to /dev/null.
A PID marker prevents a restart loop if the launcher strips the flags again;
worker threads, IPC children, unsupported platforms and an unbound addon keep
the previous warning behavior.

Instrument-hooks now initializes on first use instead of at module import, so
the discarded image cannot leave a half-initialized FIFO connection behind.

Manually verified against a rules_js 2.1.3 project (Bazel 7.4.1, Node 18.20.4)
with bazel run and bazel test: the restarted process keeps its PID, reports
the complete analysis flag set in execArgv, and stdout/stderr still reach the
test log. Outside Bazel, --test_env=CODSPEED_ENV --test_env=CODSPEED_RUNNER_MODE
is still required, since Bazel strips CODSPEED_* from tests; without it the
integration runs disabled.

Closes COD-4038

The runner mirror of getV8Flags() needs the same flag set: CodSpeedHQ/codspeed#598 adds --always-sparkplug, so the restart does not fire on the normal codspeed exec path.

[codspeed]

Merging this PR will regress 3 benchmarks

4 warnings, 1 note

⚠️ Walltime results may be unreliable

Some benchmarks used the Walltime instrument outside CodSpeed Macro Runners, which can make the results inconsistent.

Learn more about Macro Runners

See affected benchmarks

⚠️ 13 benchmarks are dominated by system calls

System calls cannot be simulated consistently and are left out of the measure. The reported value understates the real cost. Switch to the Walltime instrument to measure them.

Learn more about system calls

See affected benchmarks

⚠️ 1 benchmark with a significant change ran on different hardware

Its base and head results were measured on machines with a different CPU, core count or memory, so the change might not come from the code. Run base and head on the same runner type.

Learn more about Macro Runners

See affected benchmarks

⚠️ 12 benchmarks with a significant change ran in a different software environment

Their base and head results were measured with a different OS or toolchain, such as compiler, runtime or library versions, so the changes might come from this difference rather than from the code. Open an affected benchmark to see what changed.

See affected benchmarks

ℹ️ 16 benchmarks have no flamegraph

The flamegraph wasn't generated: most of the time was spent in runtime internals, which would leave nothing useful to analyze.

See affected benchmarks

⚡ 10 improved benchmarks
❌ 3 regressed benchmarks
✅ 257 untouched benchmarks
⏩ 56 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	Memory	wait 500ms	22.7 KB	28.7 KB	-20.98%
❌	Walltime	switch 1	72 ns	84 ns	-14.29%
❌	Walltime	switch 2	96 ns	108 ns	-11.11%
⚡	CPU Simulation	switch 1	651.7 µs	306.9 µs	×2.1
⚡	CPU Simulation	recursive fibo 10	677.4 µs	334.2 µs	×2
⚡	CPU Simulation	recursive fibo 10	676.4 µs	335.7 µs	×2
⚡	Walltime	fibo darwin	9.1 ms	5.2 ms	+75.32%
⚡	Memory	two	2 KB	1.6 KB	+24.75%
⚡	Walltime	short body	2 µs	1.7 µs	+13.79%
⚡	Walltime	short body	2 µs	1.8 µs	+13.01%
⚡	Walltime	short body	2 µs	1.8 µs	+11.56%
⚡	Walltime	switch 2	120 ns	108 ns	+11.11%
⚡	Walltime	test_iterative_fibo_10	120 ns	108 ns	+11.11%

Tip

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

---

_{Comparing cod-4038-node-introspection-restart (02e3074) with main (9c93085)}

Footnotes

1. 56 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[greptile-apps]

Confidence Score: 4/5

[Medium impact] The restart changes startup behavior in the two benchmark plugin workflows.

Fix stdin entry points before merging; they can exit successfully without running any benchmarks.

Findings

1.  Piped benchmarks do not run ▶

Fix with agent prompt

### Issue 1
packages/core/src/introspection.ts:150-155
When a benchmark is piped into Node (`node -` or Node with no script argument), Node consumes stdin before the plugin loads. If required flags are missing, this restart reuses stdin at EOF without preserving the script text. The restarted process runs an empty script and can exit successfully without running any benchmarks.

Skip the restart for stdin entry points and keep the warning behavior, or preserve and replay their source.

---

For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.

Summary

The PR restarts Node when the benchmark launcher omits required V8 flags.

• Benchmark plugins restart Node with any missing V8 flags.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart TD
  A[Plugin loads] --> B{Runner enabled and addon bound?}
  B -- No --> C[Continue without restart]
  B -- Yes --> D{Required flags missing?}
  D -- No --> C
  D -- Yes --> E{Safe context and not restarted?}
  E -- No --> C
  E -- Yes --> F[Replace Node with added flags]
  F --> G[Replay entry arguments]
  G --> H{Entry came from stdin?}
  H -- No --> I[Reload benchmark script]
  H -- Yes --> J[Read EOF and run no benchmarks]

Loading

_{Reviews (1) · Last reviewed commit: "ci: run the examples without the runner ..." · Reviewed by Greptile}

[not-matthias]

@GuillaumeLagrange lmk if you want to discuss the approach. i didn't find any real alternatives other than letting users manually pass the v8 flags which isn't really ideal imo. but considering the implementation adds complexity, we might want to consider it.