perf(gc): Date cells are movable — no longer pin old-gen defrag pages (#6186)

[proggeramlug]

Implements #6186 (Wave 3 of the 2026-07-09 GC audit). This supersedes my earlier "verified unsafe" comment on the issue — a complete moving-path enumeration showed it's actually safe; details below.

Why it's safe (the enumeration that corrected the earlier verdict)

The concern was: a moving collection running while a Date sits in an un-shadow-rooted f64 stack local would leave that local dangling. Tracing every path:

1. The movable flag only gates old-page defrag. Its sole consumers are oldgen.rs:1640/1746. The nursery copied-minor (move_young/mark_addr, copying.rs) evacuates eden objects checking only PINNED/FORWARDED — never movable. So non-movable Dates in the nursery are already evacuated at every pump-boundary moving minor today; the flag only decides whether a promoted, old-gen Date can be evacuated by defrag.
2. Old-page defrag runs only inside moving collections, i.e. gc_safepoint_moving_minor — the pump boundary (microtasks.rs, stack unwound) and, under the non-default moving-loop compiler opt-in, loop back-edges (a precise-root safepoint by construction). These are the same collections at which the nursery already evacuates Dates safely.
3. globalThis.gc() forces the conservative scan and runs a full non-moving mark-sweep (policy.rs manual_gc_collect_now), and the raw moving gc_collect_minor() is #[cfg(test)]-only. So no production moving path runs with an active JS frame holding an un-shadow-rooted Date local.

⇒ Flipping movable is safe by the identical invariant that already makes movable Promises sound — Promise is a movable arena object with the same un-shadow-rooted-local situation, and it moves fine at these safepoints.

The change

• GC_TYPE_DATE_CELL: movable: false → true, and GcMoveHookKind::None → ExoticExpandoOwner — Date's d.foo = … expandos live in object::exotic_expando keyed by address, rekeyed on relocation exactly like Promise's (exotic_expando_owner_moved).
• Temporal left non-movable — its embedded temporal_rs value + TemporalCleanup finalize hook make it a larger, separate change; scoped out.

Verification

• Unit test: a Date with an expando, evacuated by a copied minor, migrates its expando to the new address (and clears the old), preserves ts, and stays a valid Date; asserts the movable flag flipped.
• Full perry-runtime lib suite: green.
• PERRY_GC_VERIFY_EVACUATION=1: no new stale-forwarded-ref panics (only the 2 pre-existing exact-count failures that also fail on origin/main).
• Compiled smoke (default binary): 30,000 new Date() in a loop with d.seq expandos, some promoted to old-gen, a full gc() forcing moves — afterward all kept Dates are valid instanceof Date with correct getTime(), intact expandos, and working </arithmetic.

Reference: #6186; 2026-07-09 GC audit fable-audit-perry-gc.md.

Summary by CodeRabbit

• Bug Fixes

  • Improved reliability when Date values are relocated during garbage collection.
  • Preserved Date content and associated metadata after collection, preventing stale references and missed updates.

• Tests

  • Added coverage for Date relocation during minor collection to prevent regressions.

[coderabbitai]

📝 Walkthrough

Walkthrough

The GC metadata now allows Date cells to move and invokes the exotic-expando relocation hook. A regression test verifies that minor collection evacuates a rooted Date, preserves its timestamp, migrates its expando, and clears the stale address.

Changes

Date cell GC relocation

Layer / File(s)	Summary
Date cell relocation metadata and regression coverage crates/perry-runtime/src/gc/types.rs, crates/perry-runtime/src/gc/tests/copying/survival_and_malloc.rs	GC_TYPE_DATE_CELL is marked movable with ExoticExpandoOwner; the regression test verifies evacuation, timestamp preservation, expando migration, and removal from the old address.

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

Possibly related issues

• PerryTS/perry issue 6186 — The changes directly implement the Date/Temporal movability fix and add regression coverage for evacuation behavior.

Suggested labels: run-extended-tests

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly states that Date cells are now movable and mentions the GC defrag impact.
Description check	✅ Passed	The description covers summary, changes, related issue, and verification details; only optional sections are missing.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
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 docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch perf/gc-movable-dates

---

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.}

[coderabbitai]

🧹 Nitpick comments (1)

crates/perry-runtime/src/gc/types.rs (1)

469-491: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Update the Date move-hook docs Date is now movable and rekeyed via GcMoveHookKind::ExoticExpandoOwner, so the stale “Date / Temporal are non-movable” wording in crates/perry-runtime/src/object/exotic_expando.rs and the DateCell doc in crates/perry-runtime/src/date.rs should be split to keep Date separate from the still-non-movable Temporal/RegExp cases.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@crates/perry-runtime/src/gc/types.rs` around lines 469 - 491, Update the Date
move-hook documentation to reflect that Date is now movable and rekeyed by
GcMoveHookKind::ExoticExpandoOwner. In exotic_expando.rs, split the stale “Date
/ Temporal are non-movable” wording so Date is documented separately from the
still-non-movable Temporal and RegExp cases, and in date.rs update the DateCell
doc/comments to match the new movable behavior and expando rekeying.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Nitpick comments:
In `@crates/perry-runtime/src/gc/types.rs`:
- Around line 469-491: Update the Date move-hook documentation to reflect that
Date is now movable and rekeyed by GcMoveHookKind::ExoticExpandoOwner. In
exotic_expando.rs, split the stale “Date / Temporal are non-movable” wording so
Date is documented separately from the still-non-movable Temporal and RegExp
cases, and in date.rs update the DateCell doc/comments to match the new movable
behavior and expando rekeying.

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 6ae62d1a-8459-4fdb-a453-cef1b0c4b1d5

📥 Commits

Reviewing files that changed from the base of the PR and between b8cf5d3 and e47cab1.

📒 Files selected for processing (2)

• crates/perry-runtime/src/gc/tests/copying/survival_and_malloc.rs
• crates/perry-runtime/src/gc/types.rs