fix(runtime): explicit gc() runs a full collection so dead old-gen objects are reclaimed

[machineloop]

Summary

An explicit gc() did not reclaim dead old-generation / large objects, so a
large-object allocation loop grew unboundedly (RSS climbed ~175 MB/round with
gc() between rounds) instead of plateauing.

With generational GC on (the default), js_gc_collect() dispatched a minor
cycle. The minor sweep intentionally skips dead-old-block reclamation
(reclaim_dead_old_blocks = false in step_sweep), and large objects (>16 KB)
plus minor-GC survivors live in the old arena — so an explicit gc() never
returned that memory to the OS.

Changes

• crates/perry-runtime/src/gc/policy.rs — route manual_gc_collect_now (the
  single chokepoint for both the synchronous js_gc_collect path and the
  deferred flush) through gc_collect_full_mark_sweep_with_trigger, so an
  explicit gc() performs a full mark-sweep that reclaims dead old-generation
  blocks. Automatic/threshold-driven minor collections are untouched.

Related issue

n/a — explicit-gc() reclamation of old-generation/large garbage.

Test plan

# A loop allocating 20×8MB arrays/round with gc() between rounds.
# Before: RSS climbs 180 -> 355 -> 547 -> 709 (unbounded).
# After:
$ perry compile bigloop.ts -o bigloop && ./bigloop
largeloop round 1: 275 MB
largeloop round 4: 292 MB        # bounded plateau

$ cargo test -p perry-runtime --lib gc::    # 379 passed
$ cargo fmt --check -p perry-runtime         # clean

Scope / caveats (intentional)

• V8 semantics: V8's bare gc() is a scavenge (minor); gc(true) is the
  major one. This makes Perry's explicit, user-initiated gc() a thorough
  (full) collection. Gating the full sweep on gc(true) to mirror V8's
  minor/major split is a reasonable follow-up (Perry currently ignores the
  force argument).

• Pause: an explicit gc() now costs a full collection rather than a minor.
  Since gc() is user-initiated, a thorough collection is the expected trade.

• Lone large object: a single dead large object whose block is the
  allocator's current old-gen target is kept mapped + reusable (not returned
  to the OS until reused) — a deliberate allocator trade-off, separate from this
  change. This fix addresses the accumulation (unbounded-growth) case.

• cargo build --release clean — built -p perry; full-workspace build needs the GTK/gdk-pixbuf libs for perry-ui-*, which CI provides.

• cargo test --workspace --exclude perry-ui-ios --exclude perry-ui-tvos --exclude perry-ui-watchos --exclude perry-ui-gtk4 --exclude perry-ui-android --exclude perry-ui-windows passes — ran cargo test -p perry-runtime --lib gc:: (379/379) plus the end-to-end large-loop check; full workspace deferred to CI.

• (if user-facing) Added or updated a test — gc:: unit tests cover the collection paths; behavior verified end to end above.

• (if CLI / stdlib / runtime API changed) Updated docs/src/ — n/a.

• (if touching a platform UI backend) Built -p perry-ui-<backend> — n/a.

Screenshots / output

n/a — see the bounded plateau in the test plan.

Checklist

• I have NOT bumped the workspace version or edited CLAUDE.md / CHANGELOG.md (maintainer handles these at merge)
• My commits follow the loose feat: / fix: / docs: / chore: prefix convention — fix(runtime): …
• I've read CONTRIBUTING.md and agree to the Code of Conduct

Summary by CodeRabbit

• Bug Fixes
  • Explicit garbage-collection calls now perform a full collection, improving cleanup of unreachable memory.
  • Automatic, threshold-based garbage-collection behavior remains unchanged.
• Documentation
  • Added clearer notes explaining the expected behavior of manual collection calls.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: b894b521-23cb-49f1-bf6b-5b7399e26fa5

📥 Commits

Reviewing files that changed from the base of the PR and between cab9bca and 056b1a0.

📒 Files selected for processing (1)

• crates/perry-runtime/src/gc/policy.rs

---

📝 Walkthrough

Walkthrough

Explicit gc() calls now run a full mark-sweep collection with a manual trigger instead of the generational inner collection path. The added comments document the previous minor-cycle behavior and leave threshold-driven minors unchanged.

Changes

Manual GC semantics

Layer / File(s)	Summary
Manual GC uses full sweep crates/perry-runtime/src/gc/policy.rs	manual_gc_collect_now now calls gc_collect_full_mark_sweep_with_trigger(...) with GcTriggerKind::Manual, and the surrounding comments describe the explicit gc() behavior.

🎯 2 (Simple) | ⏱️ ~10 minutes

🐰 Hop-hop, a full sweep now gleams,
gc() follows its manual dreams.
Dead old blocks? Poof, away,
Little rabbit code runs clean today!

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and concisely summarizes the main runtime GC behavior change.
Description check	✅ Passed	The description follows the template well, with summary, changes, issue, test plan, output, and checklist sections filled in.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

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