bench: refresh json-polyglot and polyglot results (quiet host, 2026-08-06)

[proggeramlug]

First measurement since #7473 fixed the JSON.stringify crash that made perry produce no row at all for json_polyglot/field_access.

Conditions

Verified-quiet host — CPU active confirmed ≤25% for 60 consecutive seconds before each leg (15.2%, 15.1%, 3.7%, 4.5%), AC power, pinned Node v22.23.1 and Bun 1.3.14, 11 runs per cell, median + p95 + σ.

Headline: perry wins the JSON roundtrip

roundtrip	median	p95	σ	RSS
perry (gen-gc + lazy tape)	194 ms	197	1.5	87 MB
bun	221	235	5.1	84 MB
node	384	390	4.0	102 MB

Faster than both, with tight variance (σ 1.5).

field_access is the inverse, and the variant split is the finding

field_access	median	RSS
perry (mark-sweep, no lazy)	1340 ms	58 MB
perry (gen-gc + lazy tape)	2981 ms	218 MB
bun	223	91 MB
node	384	96 MB

The lazy tape is a 2.2× net loss on this access pattern, on ~4× the memory. That is a well-isolated optimisation target rather than a diffuse gap — the same tape that wins roundtrip loses field_access, so the decision could be per-workload.

Scope

Results files only; no code. The public artifact itself still cannot be published — #7475 (object_deep_clone / promise_all_chains fail under the auto-optimize path) is the remaining gate, and it is pre-existing rather than anything from this run.

Summary by CodeRabbit

• Documentation
  • Updated benchmark metadata to August 6, 2026.
  • Refreshed JSON polyglot and polyglot benchmark results with newer timing, memory, and per-cell measurements.
  • Updated runtime comparisons, including revised Perry results and continued unavailability of Hermes data.

[coderabbitai]

Caution

Review failed

The pull request is closed.

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 2fd936a6-03a0-4e8e-9052-bcf578468034

📥 Commits

Reviewing files that changed from the base of the PR and between a8dee4b and 8e63416.

📒 Files selected for processing (2)

• benchmarks/json_polyglot/RESULTS.md
• benchmarks/polyglot/RESULTS_AUTO.md

---

📝 Walkthrough

Walkthrough

Updated two benchmark result documents with August 6, 2026 metadata and newer measurements for runtime performance, memory usage, variability, and result ordering.

Changes

Benchmark results

Layer / File(s)	Summary
JSON polyglot benchmark measurements benchmarks/json_polyglot/RESULTS.md	Updated the benchmark date and replaced measurements for validate-and-roundtrip and parse-and-iterate workloads.
Polyglot benchmark statistics benchmarks/polyglot/RESULTS_AUTO.md	Updated benchmark metadata, headline medians, and per-cell median, p95, standard deviation, minimum, and maximum values.

Estimated code review effort: 1 (Trivial) | ~5 minutes

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch bench/results-2026-08-06

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}