docs(w47): investigation note — variance dominates the +24 % signal#84
Merged
docs(w47): investigation note — variance dominates the +24 % signal#84
Conversation
Followed the W47 thread on Mac M4 Pro this session. Stable findings: - `string_ops` is two nested loops doing `i32.div_u 10 + br_if`, classic small-hot-loop. JIT engages (interpreter is ~23x slower). - 5-run hyperfine produces σ ≈ 18 % of the mean for this specific benchmark; 20-run remeasurement on commit 9a1c76b puts: uncached: 71.8 ± 13.3 ms cached: 74.4 ± 13.5 ms vs v1.9.1 baseline of 62.2 / 64.5 ms — a uniform ~15 % slowdown on both variants. The "cached only" framing in the original W47 entry was an artifact of the 5-run sample landing on the high side; the cached vs uncached delta is noise. - A 5-run bisect across v1.9.1 → v1.10.0 (the Zig 0.15 → 0.16 migration window) would be unreliable at this variance; the measurement harness needs stabilising first. Recommendations for future sessions are in the new note. No code change; no fix attempted. Documenting the investigation so the next person to pick up W47 doesn't redo the same 30 minutes of hyperfine sampling. - `.dev/w47-investigation.md`: new file. Self-contained log: hypothesis → repro numbers → why the benchmark is noisy → next steps if/when W47 is reopened with a stable harness. - `.dev/checklist.md` W47 entry rewritten to reflect the remeasured numbers + pointer at the new note.
chaploud
added a commit
that referenced
this pull request
Apr 29, 2026
* docs: post-W50 cleanup — drop resume-guide, refocus memo on W53/C-g/W47 Plan B sub-3 (W50) and Plan C (W49) shipped via PRs #80..#83 in the 2026-04-29 PM autonomous session, plus the W47 investigation note in #84. The .dev/resume-guide.md handover doc is now stale: its "Plan B sub-3 is the next big lift" framing no longer matches reality, and the per-PR plan items are all marked complete. - Delete .dev/resume-guide.md; .dev/memo.md `## Current Task` is the single handover surface going forward. - memo.md: refresh the active-work section to W53 → C-g → W47, with the per-item plan inlined (was previously split between memo and resume-guide). - checklist.md: mark W50 done with the four-PR breakdown, refocus W49 on the C-g residual (3-platform bench baseline reset), open W53. - environment.md: explain that the only Windows-skipped CI step left is `benchmark` and link to C-g for the rationale. - roadmap.md: Windows CI guard removal flipped Done; W53 surfaced as the next active item. * fix(w53): route rustup-init stdout through Out-Host to keep return scalar PowerShell folds every native command's stdout into the enclosing function's pipeline output. Inside `Install-Rustup`, that meant rustup-init's `info: downloading component rust-std` (and the similar lines from `rustup target add wasm32-wasip1`) were piling up alongside the trailing `return $stampedDir`, so the caller's `$rustRoot = Install-Rustup ...` was a string array rather than a single path. The downstream $pathsToAdd += (Join-Path (Join-Path $paths['rust'] 'cargo') 'bin') then exploded on the empty leading element with Cannot bind argument to parameter 'Path' because it is an empty string. — matching the W53 symptom on a fresh GitHub-hosted Windows runner. Local Windows mini-PC was unaffected because rustup's "already installed" path is silent on stdout, so nothing leaked into the function's return value there. Fix: route both native command invocations through `2>&1 | Out-Host`, which keeps the lines visible in the CI log but pulls them out of the function's pipeline output. Also added a defensive check in the caller so any future regression of this shape fails loud rather than silently producing a malformed PATH. ci.yml: drop `-SkipRust` and the separate `Setup Rust` step on the Windows test job. The runner now goes through a single `install-tools.ps1` path with a self-contained `%LOCALAPPDATA%\zwasm-tools\rust-stable\` toolchain, the same as local Windows users get. * docs(w53): mark resolved with root-cause + fix summary Updates the four docs that were tracking W53 as open work to reflect the rustup-init stdout pollution diagnosis and the `Out-Host` redirect fix in the previous commit.
3 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Followed the W47 thread on Mac M4 Pro this session. Stable findings:
No code change; no fix attempted. Documenting the investigation so the next person to pick up W47 doesn't redo the same 30 minutes of hyperfine sampling.
Files
Test plan