perf(runtime): Map.clear early exit on empty maps; skip side-table probes for small numeric-keyed maps (ECS round 4)

[proggeramlug]

One runtime mechanism from the ECS round-4 chain, cut from current main. Measurement on the codehz/ecs "5k entities: 3 commands each + sync" row (paired alternating runs on the idle Mac mini) is being taken against current main and will be posted here; suites: runtime (2745) incl. map:: serial, on the isolated perrymaster gate.

• Map.clear returns early on an empty map and skips the side-tables for small numeric-keyed maps. Every index js_map_clear resets mirrors the entries exactly (insert, delete, clear and the GC rewrites keep them in step), so an already-empty map has nothing to reset — the per-entity adds.clear(); removes.clear() of an ECS change set is that case half the time. And the string and pointer side-tables hold an entry only for a string or pointer key the map holds: for a map of at most 16 entries whose keys are all bits-stable numerics (is_safe_numeric_key), reading its keys is cheaper than the two thread-local resolutions plus two hash probes that find two empty tables — which the post-4h profile showed was the whole cost of js_map_clear (both hot offsets on its two TLS probes, ~10k clears a frame on the grouping maps, 1.1% of the frame). Larger maps and any string/pointer key reset the side-tables unconditionally as before. clear_resets_every_index_whatever_the_key_kinds pins the numeric-only, empty, string-keyed and past-the-bound shapes (a stale content-hashed index entry would otherwise misroute a re-added key).

https://claude.ai/code/session_01FUvFrRNZyc5qknBiJbYbby

Summary by CodeRabbit

• Performance

  • Improved Map.clear() performance, especially for frequently cleared small maps.
  • Empty maps now clear immediately.
  • Small maps with numeric keys avoid unnecessary internal cleanup work while preserving correct behavior.

• Bug Fixes

  • Ensured all map indexes are reset correctly across numeric-keyed, string-keyed, and larger maps.

[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: d174540d-2ee4-49ee-be7b-f7673d1a270d

📥 Commits

Reviewing files that changed from the base of the PR and between 924dd16 and 2b760ec.

📒 Files selected for processing (2)

• changelog.d/8938-map-clear-fast-paths.md
• crates/perry-runtime/src/map.rs

---

📝 Walkthrough

Walkthrough

Map.clear() now returns immediately for empty maps and avoids string/pointer side-table clears for small maps with safe numeric keys. Tests cover numeric, string-keyed, empty, and large maps.

Changes

Map.clear optimization

Layer / File(s)	Summary
Implement conditional Map.clear paths crates/perry-runtime/src/map.rs, changelog.d/8938-map-clear-fast-paths.md	js_map_clear adds an empty-map fast path and skips side-table clearing for small maps whose keys are safe numerics. The changelog documents the optimization.
Validate clear behavior crates/perry-runtime/src/map.rs	Tests cover numeric-only maps, string-keyed maps, empty maps, and maps above the scan limit.

Estimated code review effort: 2 (Simple) | ~10 minutes

Suggested reviewers: thehypnoo

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

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.}

[proggeramlug]

Merged. Rebased onto current main; it conflicted with #8934 in map.rs because both add a test at the same spot and the two bodies share one closing brace. Kept both tests, each with its own brace and #[test].

Validation: runtime 2770/0 (RUST_TEST_THREADS=1), scripts/run_lint_gates.sh 57 of 58 including the compile tier green — the exception is the pre-existing ${{ }} substitution artifact (#8929).

[proggeramlug]

Paired measurement (idle Mac mini, 9 alternating pairs, codehz/ecs "5k entities: 3 commands each + sync"): control = main 49279e99e, candidate = this branch on that base: 3.955 → 3.943 ms/op, +0.36%, 9/9 (r4j-screen.json, process oracles 18/18). Small, unanimous.