Bench: cover Bun isolated .bun store; fix rescan restore

[Mikola Lysenko (mikolalysenko)]

What main changed

#496 (35de7548, "Fix npm crawler missing Bun, Deno and Yarn 4 stores") made the npm crawler walk Bun's isolated store, node_modules/.bun/<name>@<version>/node_modules/<name>, in scan, apply's resolver and VEX. Bun 1.3.2+ uses the isolated linker by default. The suite only had a hoisted Bun project (bun/*), so this path was never timed or validated.

Suite changes

• Added bun-isolated/hosted and bun-isolated/rescan (3000 packages, 60 patched). They use the same text bun.lock generator as bun/*, with a pnpm-shaped .bun store: per-entry dependency symlinks, Bun's .bun/node_modules hoist links, and root links for direct dependencies only, so 2700 of the 3000 packages exist only in the store. The Expect matches bun/*: every package scanned, 60 redirected, bun.lock rewritten.
• Updated the README's package-manager list.
• Fixed (93b1c929, harness): when a rescan's preparatory scan failed validation, run_once returned before restoring the tree. In compare the other binary then started from a rescanned project and was reported INVALID by mistake. The 2026-10-04 weekly A/B hit this: the pre-Fix npm crawler missing Bun, Deno and Yarn 4 stores (#366, #373, #405, #495) #496 base failed bun-isolated/rescan's prep, and the head, which validates on its own, was flagged redirect.rewrittenFiles: got [], want ["bun.lock"]. The harness now restores before returning the error.
• Removed: nothing.

Validation (4 vCPU Intel Xeon @ 2.80GHz cloud sandbox)

• run on main 045d7ec7, 3 runs: bun-isolated/hosted 321.9 ms, bun-isolated/rescan 332.6 ms, 127 requests, 34.8 MiB peak RSS. Both validate.
• The scenario exercises the new path: the pre-Fix npm crawler missing Bun, Deno and Yarn 4 stores (#366, #373, #405, #495) #496 binary (1169ae68) fails it with lockfileOnlyPackages: got 2700, want 0.
• A/A compare (default 15 pairs): hosted +0.8% [−2.1, +6.1], rescan −0.7% [−4.0, +6.4]. No regression.
• strace -f -e trace=execve on a serve run shows only env → socket-patch, so no subprocesses.
• cargo fmt -p socket-patch-bench, cargo clippy -p socket-patch-bench --all-features --all-targets -D warnings and cargo test -p socket-patch-bench (29 passed) are all clean.
• Harness fix (2026-10-04, 4 vCPU Xeon @ 2.10GHz): re-running 2463257a vs main on bun-isolated/rescan + yarn-berry/rescan now reports only the base invalid. A/A, 7 pairs: npm/rescan +1.2% [−10.2, +13.9], bun-isolated/rescan +1.2% [−7.9, +12.2], poetry/rescan −4.8% [−11.0, +12.3]. No regressions. fmt, clippy -D warnings and tests (29 passed) are clean.

Time budget

The harness fix doesn't change timing (restore runs only on a failed prep). A default (15-pair) compare of the two new scenarios took 89 s on this sandbox, fixture builds included. On this machine a 9-pair compare of the existing 39 scenarios took 8.4 min. GitHub's 4-core runners have run faster than this sandbox so far, so a full default compare should stay near the ~12 min target. If it goes over, bun/rescan is the first candidate to drop, since its code path is now covered twice.

🤖 Generated with Claude Code

https://claude.ai/code/session_016Gubp3gsnxFDvRbguaeLqT

---

Note

Low Risk
Changes only the benchmark harness and fixtures; the rescan restore fix improves test isolation with no production runtime impact.

Overview
Adds bun-isolated hosted/rescan scenarios to socket-patch-bench so timing and validation cover Bun 1.3’s default isolated node_modules/.bun store (same bun.lock as hoisted bun, but a pnpm-shaped symlink layout with most packages store-only). The README now separates hoisted bun from bun-isolated.

Fixes a harness bug in run_once: when a rescan’s preparatory scan fails validation, the work tree is restored before returning the error so A/B compare does not leave the next binary starting from a partially rewritten fixture and failing validation incorrectly.

^{Reviewed by Cursor Bugbot for commit 93b1c92. Configure here.}

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

bugbot run

---

Generated by Claude Code

[cursor]

Stale Bugbot comment from a previous run.

[Mikola Lysenko (mikolalysenko)]

Ready for review at 5ab8e870603f892b842bd45f4fcb5f54597145a7.

• CI: 97/97 green (3 skipped).
• Bugbot: reviewed this head, no findings.
• Reviewer focus: bench-only change (new bun-isolated fixture); check the default compare time stays near the ~12 min budget.

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

Codex review of 5ab8e870603f892b842bd45f4fcb5f54597145a7: ready to merge as-is from this review. No actionable findings.

Reviewed all three changed files, fixture registration, dependency/version resolution, and scoped/unscoped symlink targets. Validation passed:

• 29 repository tests in the benchmark crate and benchmark Clippy.
• Both full-size bun-isolated/hosted and bun-isolated/rescan scenarios against the exact-head CLI: 3,000 packages and 60 patches, with all correctness assertions satisfied.
• Full fixture topology: 300 direct root links, 2,700 instances without root links, and all 9,099 symlinks valid; dependency targets match the declared package names and versions.
• A native Bun 1.3.2 isolated install confirms scoped and unscoped store keys, root/dependency/hoist links, and duplicate-version placement.

Full current-head CI is clear: 245 successful checks, six skipped; six successful workflows, two skipped. Exact-head Bugbot is clear, with no unresolved review threads. The unchanged source merges cleanly with current main 045d7ec7.

The two debug scenario runs validate correctness; they do not independently establish the author's performance comparison. No source changes were needed.

[Mikola Lysenko (mikolalysenko)]

bugbot run

---

Generated by Claude Code

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 93b1c92. Configure here.}

[Mikola Lysenko (mikolalysenko)]

[burn-down agent] Still Ready for review, now at 93b1c9295b938365c30be564fb21d1f7b35de1b4. The label was first applied at 5ab8e87.

• CI: 249/249 check runs green on this head (4 skipped by path filters), none failing.
• Bugbot: reviewed 93b1c92, no new issues. No unresolved review threads. Mergeable and up to date with main.
• New since the last ready comment: 93b1c92 restores the bench fixture after a failed rescan prep, so a dirty fixture no longer leaks into the next binary's run. Reviewers should check the restore path in the rescan prep.

---

Generated by Claude Code