[tarushmohindru]

This Pull Request is a part of #5567.

It changes the following:

• Adds a gc-stats feature to boa_gc, forwarded by boa_engine and boa_cli. It is off by default and gates all new code.
• Records per-thread statistics in BoaGc: collections, collections that ran finalize + second mark, cumulative bytes/objects allocated and freed, and wall time per collection phase plus total.
• Adds boa_gc::stats() -> GcStats (a Copy snapshot) and boa_gc::reset_stats().
• Adds $boa.gc.stats() and $boa.gc.resetStats() to the CLI debug object.
• Adds tests in core/gc/src/test/stats.rs.

Design notes

• Everything is behind the feature. Without gc-stats the collector is unchanged. std::time::Instant::now() panics on wasm32-unknown-unknown, so timing has to stay opt-in.
• Separate from GcRuntimeData. Stats live in their own stats: GcStats field on BoaGc, so reset_stats() can't touch bytes_allocated or the collection threshold. runtime.collections is left as is (tests rely on it), and GcStats::collections is a separate counter that resets.
• Shallow byte counts. Only size_of_val of each GcBox/EphemeronBox is counted, matching the existing bytes_allocated. Heap memory owned by values (e.g. a Vec's buffer) and the boxes backing WeakMaps are not counted. This is documented on GcStats.
• Freed counts without touching sweep. Bytes and objects freed are computed from bytes_allocated and the heap lengths before and after the sweep, so the unsafe code is unchanged.
• Phase timing uses a small private timed! macro that expands to the plain expression when the feature is off, which keeps collect() readable.
• GcStats is #[non_exhaustive] so more counters can be added later without a breaking change.

Overhead

Release boa_cli built with and without --features gc-stats; V8 Richards and Splay from benches/scripts/v8-benches, 10 runs each, alternating builds. AMD Ryzen 5 5600H, WSL2, based on 9871775a. Higher is better; median with range.

Suite	Off	On	Δ
Richards	233 (228–236)	230 (221–231)	−1.3%
Splay	895.5 (877–901)	911.5 (886–921)	+1.8%

Both differences are within code-layout noise: Splay is faster with stats enabled, and Richards performs zero collections and about 8k allocations in total, so the counters can't account for a 1.3% change. Happy for someone with a quieter setup to re-run these.

Example: Splay

With --features gc-stats --debug-object, printing $boa.gc.stats() after the run:

Collections	18 (all 18 ran finalize + second mark)
GC time	~970 ms of ~3.1 s wall time (~31%)
mark	574 ms
trace_non_roots	177 ms
sweep	130 ms
remark	55 ms
finalize	44 ms

This lines up with the perf estimate in #5567 (~36% of samples), and gives a direct data point for the Phase 1 NEEDS_FINALIZE discussion: every collection on Splay paid for finalize and a second full mark.

Testing

• cargo test -p boa_gc and cargo test -p boa_gc --features gc-stats
• cargo clippy --all-targets for boa_gc and boa_cli, with and without the feature

[github-actions]

Test262 conformance changes

Test result	main count	PR count	difference
Total	53,578	53,578	0
Passed	51,450	51,450	0
Ignored	1,648	1,648	0
Failed	480	480	0
Panics	0	0	0
Conformance	96.03%	96.03%	0.00%

Tested main commit: 9871775a4999591e27bbb026b92530bf43caef52
Tested PR commit: 1b4c8189130fb77ceeb85b8f4dd9ff6d3cb4d787
Compare commits: 9871775...1b4c818

[codecov]

Codecov Report

❌ Patch coverage is 66.66667% with 11 lines in your changes missing coverage. Please review.
✅ Project coverage is 63.09%. Comparing base (6ddc2b4) to head (1b4c818).
⚠️ Report is 1083 commits behind head on main.

Files with missing lines	Patch %	Lines
core/gc/src/lib.rs	78.57%	6 Missing ⚠️
cli/src/debug/gc.rs	0.00%	5 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##             main    #5613       +/-   ##
===========================================
+ Coverage   47.24%   63.09%   +15.85%     
===========================================
  Files         476      536       +60     
  Lines       46892    60553    +13661     
===========================================
+ Hits        22154    38208    +16054     
+ Misses      24738    22345     -2393     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[jasonwilliams]

Sorry this took a while to look at

LGTM