fix(gc-tests): suppress automatic GC triggers around raw-pointer forwarding setup

[jdalton]

Summary

prototype_addr_cache.rs and side_table_scanners.rs (both under
crates/perry-runtime/src/gc/tests/runtime_roots/) hand-stage GC forwarding
chains by allocating with arena_alloc_gc / arena_alloc_gc_old and holding
the raw returned pointer/address in a native local — with no root
registered — while further allocations happen. Those further allocations can
land on the block-full slow path (arena_cell_alloc), which calls
gc_check_trigger(). Without suppression, an automatic collection there can
relocate or free the held object out from under the test before it finishes
staging the forwarding chain.

Neither file used the guard its siblings (callback_scanners.rs,
hook_dispatch_handles.rs, transient_handles.rs) already rely on for this
exact hazard: GcTriggerThresholdTestGuard::suppress_automatic_triggers().
This PR adds it at every exposed call site in both files.

Enumerated call sites

prototype_addr_cache.rs

Site	Live raw pointer across the call	Disposition
nursery_stand_in() — arena_alloc_gc	none (first alloc in the function)	Not exposed — comment added
evacuate() — arena_alloc_gc_old	from	Guarded
forwarded_pair() — via evacuate()	from, live in this frame across the call into evacuate	Guarded (defense-in-depth alongside evacuate's own guard)
prototype_addr_reads_through_a_forwarding_stub (test)	none directly — forwarded_pair() already returns plain usizes and carries its own guard	Not exposed — comment added
prototype_addr_reads_through_a_multi_hop_forwarding_chain (test) — 2nd arena_alloc_gc, then arena_alloc_gc_old	first live across the 2nd arena_alloc_gc; first + second live across arena_alloc_gc_old	Guarded
prototype_addr_cache_is_rewritten_by_the_collector (test) — 2nd nursery_stand_in(), then two evacuate() calls	array_from live across the 2nd nursery_stand_in() call	Guarded (found during audit — not in the original hazard list)
prototype_addr_cache_scanner_is_registered, prototype_addr_cache_scanner_leaves_the_unset_sentinel_alone	no arena_alloc_gc* calls	Not applicable

side_table_scanners.rs

Test	Exposed call(s)	Disposition
test_implicit_this_root_scanner_marks_and_rewrites	nursery_user live across arena_alloc_gc_old	Guarded
test_class_side_table_scanner_marks_values_but_not_function_keys	dynamic_value/prototype_value/cached_value/prototype_object live across both arena_alloc_gc calls (parent_closure, function_key); parent_closure live across function_key's call	Guarded
test_registered_class_side_table_scanner_rewrites_values_and_function_keys	value_user live across key_user's arena_alloc_gc; both (+ value_old) live across the two arena_alloc_gc_old calls	Guarded
test_symbol_side_table_scanner_marks_keys_and_values_without_marking_owner	none — every allocation goes through opaque helpers (js_object_alloc, alloc_nursery_test_symbol, young_leaf) with no literal arena_alloc_gc* call written in this file	Not exposed — comment added
test_symbol_side_table_registered_scanner_rewrites_roots_and_metadata	owner/sym_key/value/static_sym_key/static_value (+ _old addresses) live across the three arena_alloc_gc_old calls	Guarded
test_runtime_root_visitor_rewrites_raw_pointer_slots	nursery_user live across arena_alloc_gc_old	Guarded
test_class_inheritance_side_table_roots_mark_and_rewrite	proto_user/decl_proto_user/closure_user (+ _old addresses) live across the later arena_alloc_gc/arena_alloc_gc_old calls	Guarded
test_runtime_root_visitor_rewrites_cell_and_atomic_slots	nursery_user live across arena_alloc_gc_old	Guarded
test_runtime_root_visitor_rewrites_metadata_without_marking	nursery_user live across arena_alloc_gc_old	Guarded

Note: young_leaf, alloc_nursery_test_symbol, alloc_old_test_symbol, and
js_object_alloc (shared helpers, mostly in support.rs) have the same
underlying shape but are out of scope here — they're used far beyond these
two files, and this PR is scoped to the two files' own literal
arena_alloc_gc/arena_alloc_gc_old call sites per the task that opened it.
Flagging for awareness, not fixing inline.

Test plan

• cargo test -p perry-runtime runtime_roots — 77 passed, 0 failed
• cargo test -p perry-runtime --lib -- --test-threads=1 on this branch: 1645 passed, 0 failed, 3 ignored
• Same single-threaded full run on the parent commit (97c6921): 1645 passed, 0 failed, 3 ignored — identical failing set (empty)
• Default (multi-threaded) full runs show a handful of order-dependent flakes on both the parent and this branch (different tests each time — gc::tests::teardown, pty, object::prop_plan, object::global_this_webassembly), confirming they're pre-existing test-isolation flakiness unrelated to this change, not a regression
• cargo fmt --check -p perry-runtime clean

Summary by CodeRabbit

• Tests

  • Improved garbage collection test reliability by preventing unintended automatic collections during allocation-sensitive scenarios.
  • Added safeguards and explanatory comments for tests that retain objects across allocations.
  • Clarified coverage for pointer lifetime and object movement behavior.

• Documentation

  • Documented garbage collection test protections and the reasoning behind guarded and unguarded allocation cases.

[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: 6bed74a4-06ec-4afe-9f12-9d535e73603b

📥 Commits

Reviewing files that changed from the base of the PR and between c109480 and 73d8d68.

📒 Files selected for processing (3)

• changelog.d/7283-gc-test-trigger-suppression.md
• crates/perry-runtime/src/gc/tests/runtime_roots/prototype_addr_cache.rs
• crates/perry-runtime/src/gc/tests/runtime_roots/side_table_scanners.rs

🚧 Files skipped from review as they are similar to previous changes (2)

• crates/perry-runtime/src/gc/tests/runtime_roots/prototype_addr_cache.rs
• crates/perry-runtime/src/gc/tests/runtime_roots/side_table_scanners.rs

---

📝 Walkthrough

Walkthrough

The changes add automatic-GC trigger suppression guards and pointer-liveness comments to runtime-root tests. The changelog documents the guarded allocation cases and the opaque-helper exception.

Changes

GC test safety

Layer / File(s)	Summary
Prototype forwarding guards crates/perry-runtime/src/gc/tests/runtime_roots/prototype_addr_cache.rs	Evacuation, forwarding, multi-hop forwarding, and collector rewrite tests suppress automatic triggers while raw pointers remain live. Comments document the guard requirements.
Side-table allocation guards crates/perry-runtime/src/gc/tests/runtime_roots/side_table_scanners.rs	Side-table scanner setup paths add trigger-suppression guards around direct allocations. Comments document live values and the opaque-helper case without a guard.
GC suppression changelog changelog.d/7283-gc-test-trigger-suppression.md	Documents the new suppression guards and pointer-liveness assumptions.

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

Possibly related PRs

• PerryTS/perry#7050: Changes arena allocation and GC-trigger behavior in related runtime code.
• PerryTS/perry#7071: Introduces the prototype address-cache tests updated by this PR.
• PerryTS/perry#7196: Addresses stale raw GC pointers across allocations in other runtime-root tests.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and concisely describes the main change: suppressing automatic GC triggers around raw-pointer forwarding setup in GC tests.
Description check	✅ Passed	The description clearly explains the hazard, enumerates affected call sites, documents scope, and provides detailed test results.
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 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.}