[not-matthias]

In b.Loop() mode, each iteration is saved as one round. The round time was the time since the last StartTimer call. If an iteration stops and restarts the timer, the round drops every timed segment before the last StartTimer:

for b.Loop() {
	work() // timed, but dropped from the round

	b.StopTimer()
	teardown()
	b.StartTimer() // the round starts here
}

The uploaded walltime results then only contain the few hundred nanoseconds between that StartTimer and the next b.Loop(). The ns/op that go test prints was correct because it uses b.duration, so the two numbers did not match.

SaveMeasurement now saves the timed duration accumulated since the previous round (b.duration - b.savedDuration). b.N rounds are unchanged: runN resets the timer before each round, so the delta is b.duration.

Local repro with 1 ms of timed work and untimed setup and teardown in each iteration:

Benchmark	go test ns/op	Uploaded median before	Uploaded median after
b.Loop() with StopTimer/StartTimer	1,000,289	111	1,000,906
b.Loop() without timer calls	1,000,182	1,000,195	1,000,134

BenchmarkWithTeardownInLoopModern in example/ covers this pattern. test_b_loop_round_includes_work_before_stop_timer checks that its round includes the 1 ms of timed work. Without the fix, the test fails with a round of 11.8 µs.

Closes COD-4085

[greptile-apps]

Confidence Score: 4/5

[Medium impact] The change affects the benchmark rounds CodSpeed records for Go runs.

Fix the new test’s missing Go-version check before merging.

Findings

1.  Go 1.24 CI fails ▶

Fix with agent prompt

### Issue 1
go-runner/src/integration_tests.rs:117
The new test runs an example that requires Go 1.25 without the `can_build_project` check used by `test_build_and_run`. The runner creates its overlay against the installed Go tree before running the example. If Go switches to 1.25 to satisfy `example/go.mod`, those paths no longer match, so the test gets no CodSpeed results and fails.

Add the same version check before calling `run_benchmarks`.

---

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

Summary

The PR makes each b.Loop() round include all timed segments, rather than only the last segment.

• Each b.Loop() round includes all of its timed work.

_{Reviews (1) · Last reviewed commit: "fix(go-runner): count all timed segments..." · Reviewed by Greptile}

[codspeed]

Merging this PR will improve performance by 19.05%

2 warnings

⚠️ 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

⚠️ Unable to measure the execution of 1 benchmark

This benchmark measured no execution time, so its result is not comparable and counts as unchanged. This usually happens when the compiler removes the code under test.

Prevent compiler optimizations

See affected benchmarks

⚡ 1 improved benchmark
✅ 48 untouched benchmarks
🆕 1 new benchmark
⏩ 1 skipped benchmark¹

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
⚡	BenchmarkSleep100ns_Loop	300 ns	252 ns	+19.05%
🆕	BenchmarkWithTeardownInLoopModern	N/A	1.1 ms	N/A

Tip

Curious why performance improved? Comment @codspeedbot explain why performance improved on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing cod-4085-investigate-suspicious-go-walltime-results-with-timer (5a63029) with main (a5f6b7c)}

Footnotes

1. 1 benchmark was skipped, so the baseline result was used instead. If it was deleted from the codebase, click here and archive it to remove it from the performance reports. ↩