[cknitt]

The benchmarks job failed on #9267 (run) with build/leaf-edit +348% and build/no-op +54% instructions, although their allocated words didn't change by 0.1%.

The cause: rescript build runs with an interrupt poll, which starts a ticker thread (Process_child.with_completion_notifier) that wakes the scheduler every 5 ms of wall-clock time while work runs. Under cachegrind the build is 20–50× slower, so the number of wake-ups, and the instructions they execute, depend on how fast the runner is. #9272 had seen 9.5% on the shared-edit build already and kept the two small builds, which then happened to match.

This stops counting instructions for the project builds; their allocated words, which agree within 0.0004% between runs, stay. The replayed single-file builds, stress inputs and parse/format benchmarks keep their instruction counts. The job also gets faster by the two cachegrind builds per side.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Kbf3gGPsF3fHZn9rBxA7Hm

[github-actions]

Changelog

No changelog entry.

Semantic Diff

2 commits · Full page

Stop counting instructions for the project builds

_{8eb8855 · cknitt + Claude · Nov 09, 2026, 09:29 UTC}

A ticker wakes the build's scheduler every 5 ms of wall-clock time, so under cachegrind the builds' instructions depend on how fast the runner is: in #9267 the leaf-edit build grew by 348% and the no-op build by 54% although their allocations didn't change, failing the job. Their allocated words stay, and the job runs faster.

✨ Retrigger the DCO check

_{067d97f · cknitt + Claude · Nov 09, 2026, 11:11 UTC · new}

The DCO app left its check suite for the previous commit queued and never reported; a new commit makes it check the pull request again.

[github-actions]

✅ Compiler benchmarks

Allocated words

In total: 806,109,965 → 806,109,889 (±0.00%).

None of the 261 benchmarks changed by 0.1% or more.

Instructions

In total: 10,226,962,100 → 10,226,962,100 (±0.00%).

None of the 33 benchmarks changed by 0.1% or more.

Output size

In total: 29,806,856 → 29,806,856 (±0.00%).

None of the 13 benchmarks changed by 0.1% or more.

Pull request 067d97f compared with base bcb7ec7 in this run.

[github-actions]

Developer playground preview: https://rescript-lang.github.io/rescript/dev-playground/?version=pr-9274

[pkg-pr-new]

Open in StackBlitz

rescript

npm i https://pkg.pr.new/rescript@9274

@rescript/belt

npm i https://pkg.pr.new/@rescript/belt@9274

@rescript/darwin-arm64

npm i https://pkg.pr.new/@rescript/darwin-arm64@9274

@rescript/linux-arm64

npm i https://pkg.pr.new/@rescript/linux-arm64@9274

@rescript/linux-x64

npm i https://pkg.pr.new/@rescript/linux-x64@9274

@rescript/runtime

npm i https://pkg.pr.new/@rescript/runtime@9274

@rescript/win32-x64

npm i https://pkg.pr.new/@rescript/win32-x64@9274

commit: 067d97f