perf(runtime): memoize call/apply rest dispatch by proggeramlug · Pull Request #10127 · PerryTS/perry

proggeramlug · 2026-09-12T12:13:40Z

Summary

route lookup_closure_rest_full through the existing four-entry dispatch-strategy memo instead of probing CLOSURE_BODY_REGISTRY on every call-array dispatch
retain the memo's registration-time invalidation, including the call-before-rest-registration case from methods inherited via Object.setPrototypeOf(obj, proto) run with this=undefined — effect Pipeable/Tag statics return the wrong pipe stage (blocks web.ts, 'Not a valid effect: undefined') #6475
add a bounded counter regression: 32 calls to the same non-rest closure perform one strategy miss and one total body-registry lookup
cover rest and non-rest arities 0–16 across direct/spread, .call, and .apply, plus ordinary/arrow/bound receivers and the imported-class 0x7FFE carve-out

Performance

Measured from clean release builds of current main (50e08e91dd6a54d9d9210c43a5d36c86d880d144) and this patch on the same AMD Ryzen 7 7700X host. Both compilers and matching static runtime/stdlib archives were rebuilt with cargo build --release --locked; reproducers used --no-auto-optimize. Node was exact v26.5.1 (SHA-256 fb48e77df2f8e92fedfec39afa60a5f41563441f6b61316ada5fb295a431c2c6). Runs were serialized on one pinned CPU, with baseline/candidate order alternated. Values below are medians across paired complete sweeps; ratio is Perry/Node.

`function-call` (5 paired sweeps)

n	main ms	candidate ms	main ratio	candidate ratio	ratio improvement	checksum
100	0.021941	0.021685	82.426x	81.545x	1.07%	53206851
1,000	0.219205	0.215381	88.821x	87.262x	1.76%	509020824
10,000	2.190220	2.151382	89.439x	87.754x	1.88%	6312813
100,000	21.847207	21.541055	89.323x	87.955x	1.53%	65781663
1,000,000	219.758416	216.032604	40.484x	39.727x	1.87%	468838303

`function-apply` (7 paired sweeps)

n	main ms	candidate ms	main ratio	candidate ratio	ratio improvement	checksum
100	0.027578	0.027602	104.478x	104.331x	0.14%	53206851
1,000	0.275673	0.275316	112.354x	111.981x	0.33%	509020824
10,000	2.782989	2.753682	114.230x	113.278x	0.83%	6312813
100,000	27.709004	27.521415	113.992x	113.453x	0.47%	65781663
1,000,000	277.215106	275.442614	51.099x	50.862x	0.46%	468838303

All Node/Perry checksums matched. Every paired median ratio improved, but the sub-1% movements—especially apply at n=100—are noise-sensitive. The deterministic counter test is the evidence for the targeted result: after the first resolution, repeated dispatch performs no closure-body hash lookup. The remaining gap is dominated by the wider method-call/apply path, consistent with the issue's note that its source attribution was not exclusive.

Validation

cargo test -p perry-runtime dispatch_recent_tests --lib
cargo test -p perry-runtime closure:: --lib -- --test-threads=1
cargo test -p perry-runtime body_record_tests --lib
cargo fmt --all --check
cargo clippy -p perry-runtime --profile perry-dev
python3 scripts/check_test_registration.py
release Perry and exact Node v26.5.1 both print ok for test_gap_10085_call_apply_dispatch.ts

The full parallel runtime suite reported three state-sensitive failures; each passed immediately in isolation. pre-tag-check --quick passes all relevant checks but reports the pre-existing public benchmark-evidence freshness failure on unmodified benchmark/Cargo inputs.

Coordination

The sibling #10084 / PR #10119 branch does not touch registry.rs or this memo, and this patch does not touch bind metadata.

Summary by CodeRabbit

Performance
- Improved repeated function calls using call, apply, and spread syntax by reducing dispatch overhead.
- Preserved behavior for rest parameters, argument counts, receivers, bound functions, and imported callable classes.
Bug Fixes
- Fixed dispatch behavior when rest-parameter functions are registered after earlier calls.
Tests
- Added coverage for cached dispatch reuse and regression scenarios across supported invocation patterns.

coderabbitai · 2026-09-12T12:14:14Z

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Advanced

Run ID: 5c2c985c-1ad4-4bfd-9428-ed0a6b6b8cfc

📥 Commits

Reviewing files that changed from the base of the PR and between 50e08e9 and 0d6eb03.

📒 Files selected for processing (4)

changelog.d/10127-call-apply-dispatch-cache.md
crates/perry-runtime/src/closure/registry.rs
test-files/fixtures/issue_10085_dispatch/imported.ts
test-files/test_gap_10085_call_apply_dispatch.ts

Included review availability: Your plan provides up to 8 included reviews per hour; 4 remain after this review.

📝 Walkthrough

Walkthrough

The runtime now resolves closure rest metadata through the existing dispatch memo. Tests verify cache reuse, late-registration invalidation, call/apply/spread semantics, receiver handling, argument padding, and imported class references.

Changes

Closure dispatch cache

Layer / File(s)	Summary
Memoized rest dispatch `crates/perry-runtime/src/closure/registry.rs`, `changelog.d/10127-call-apply-dispatch-cache.md`	`lookup_closure_rest_full` uses cached dispatch resolution. Test instrumentation verifies lookup reuse and invalidation after rest registration.
Call and apply semantic coverage `test-files/fixtures/issue_10085_dispatch/imported.ts`, `test-files/test_gap_10085_call_apply_dispatch.ts`	Tests cover rest arguments, missing parameters, receiver behavior, bound calls, and imported class references across direct, `.call`, and `.apply` invocation.

Priority: ➖ Normal

Estimated code review effort: 3 (Moderate) | ~25 minutes

Change: Refactor · Severity of issue fixed: Medium

Merge Risk: ⚪ Minimal · up to 0d6eb

The dispatch memoization retains late-registration invalidation and is covered across call, apply, spread, receiver, arity, and imported-class scenarios. No merge-blocking risk remains.

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 42.86% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 14 functions across 3 files. (1 skipped: …	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and concisely describes the main change: memoizing call/apply rest dispatch in the runtime.
Description check	✅ Passed	The description is detailed and relevant. It covers the change, issue, performance results, validation commands, test coverage, and coordination. It does not reproduce every template heading, and it o…
Linked Issues check	✅ Passed	The PR satisfies the coding requirements in [`#10085`]. `lookup_closure_rest_full` now uses the existing four-entry dispatch memo instead of calling `body_record` on each call-array dispatch. Registrati…
Out of Scope Changes check	✅ Passed	The changes stay within [`#10085`]. Registry changes implement dispatch caching and invalidation. The Rust and TypeScript tests verify the required dispatch, receiver, arity, rest, registration, and cla…

Full details: Docstring Coverage

Explanation

Docstring coverage is 42.86% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 14 functions across 3 files. (1 skipped: 1 unsupported.)

Fix all pre-merge checks with AI

✨ 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

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

Train167 (#10115, #10125, #10126, #10127, #10133, #10134) lands on main at 0.5.1539; none of the PRs bumped the version, which is the maintainer's job at merge time. Cargo.lock regenerated so every workspace member's inherited version moves with it.

`gc_runtime_root_holders` failed on the train: the call/apply rest-dispatch cache adds a new identity-ratcheted thread-local that nothing classifies. crates/perry-runtime/src/closure/registry.rs:312: BODY_RECORD_LOOKUPS: std::cell::Cell<u32> [rule T] It is a counter inside a `#[cfg(test)]` `std::thread_local!` block — absent from production builds, holding a `u32` count and never an address — so it takes a researched `test_only` verdict. Its sibling in the same block, RESOLVE_STRATEGY_SLOW_CALLS, is pinned on the frontier as debt. A pinned entry is explicitly not a GC-safety verdict, so the new one is classified rather than pinned; that is the stronger record and it keeps the frontier list from growing for a value that is trivially provable.

proggeramlug · 2026-09-12T15:51:53Z

Landed on main via merge train #10139 (rebase-merged, per-commit authorship preserved).

Your commits are on main starting at 46120666c5; the train tree was verified identical to main after the merge (git diff origin/main HEAD --stat empty).

One maintainer follow-up: gc_runtime_root_holders rejected the new BODY_RECORD_LOOKUPS thread-local as unclassified. It is a Cell<u32> in a #[cfg(test)] block, so I gave it a researched test_only verdict rather than pinning it on the frontier as its sibling RESOLVE_STRATEGY_SLOW_CALLS is — a pinned entry is explicitly not a GC-safety verdict, and this one is trivially provable.

Closing as landed — GitHub cannot auto-close through a train branch.

Ralph Küpper added 2 commits September 12, 2026 14:10

perf(runtime): cache call/apply rest dispatch

0e76900

chore: number changelog fragment for PR 10127

0d6eb03

proggeramlug mentioned this pull request Sep 12, 2026

merge train: land #10115, #10125-#10135 (twelve PRs) #10139

Merged

proggeramlug closed this Sep 12, 2026

This was referenced Sep 12, 2026

perf(runtime): every .call/.apply pays a thread-local registry hash lookup, costing a flat 58x/68x Node #10085

Closed

perf: reduce startup overhead and specialize proven tiny programs #10066

Open

Navigation Menu

Uh oh!

Uh oh!

perf(runtime): memoize call/apply rest dispatch - #10127

perf(runtime): memoize call/apply rest dispatch#10127
proggeramlug wants to merge 2 commits into
PerryTS:mainfrom
proggeramlug:perf/10085-call-apply-dispatch

proggeramlug commented Sep 12, 2026 •

edited by coderabbitai Bot

Loading

Uh oh!

coderabbitai Bot commented Sep 12, 2026 •

edited

Loading

Walkthrough

Changes

❌ Failed checks (1 warning)

Uh oh!

proggeramlug commented Sep 12, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Uh oh!

Uh oh!

Conversation

proggeramlug commented Sep 12, 2026 • edited by coderabbitai Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Performance

function-call (5 paired sweeps)

function-apply (7 paired sweeps)

Validation

Coordination

Summary by CodeRabbit

Uh oh!

coderabbitai Bot commented Sep 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Walkthrough

Changes

❌ Failed checks (1 warning)

Uh oh!

proggeramlug commented Sep 12, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

proggeramlug commented Sep 12, 2026 •

edited by coderabbitai Bot

Loading

`function-call` (5 paired sweeps)

`function-apply` (7 paired sweeps)

coderabbitai Bot commented Sep 12, 2026 •

edited

Loading