fix(runtime): match node on the gap-suite regressions

[steinybot]

Merge in this order:

1. fix(docs): the my-bindings fixture reads buffers inside a no-GC scope #12112
2. fix(ext): ext-crate tests root values through RuntimeHandleScope #12113
3. fix(runtime): audit the new raw stores and classify addresses with addr_class #12114
4. ci: register the dark GC witnesses and isolate the new test counters #12115
5. refactor: split five files back under the 2000-line cap #12116
6. fix(runtime): root raw handles across calls and correct the handle ratchets #12117
7. fix(runtime): instanceof never mints a class object for a builtin id #12118
8. fix(gc_effects): exempt aHash's random-source init from the indirect-call taint #12119
9. docs: regenerate the API reference #12120
10. fix(runtime): clear the gates that #12109 left red #12121
11. fix(runtime): match node on the gap-suite regressions #12149 (this PR)

Problem

CI on main is red again after #12025. Commits since then broke check, warnings, cargo-test, gc-call-effects, self-test-checkers, gap-suite and fourteen lint steps, so no PR can show a green build. Each later failure in a job was hidden behind an earlier one.

Solution

Fix each breakage in its own PR. Each PR targets main and holds every fix above it in the order list, so the last PR is the one that can run fully green. Merge in order: each diff then shrinks to its own fix.

In this PR

Fixes the gap-suite regressions. Main's gap shards are usually cancelled by the next push, so about 30 tests had been failing unseen. None comes from this chain: base and tip give identical results. This PR also merges current main and fixes the gates that main's newest commits left red. Main has since fixed some of the regressions itself, and this PR takes main's fix in each case.

Changes

1. Brand-check the Function.prototype call and apply thunks (diff). Once lowering stopped folding apply.call(undefined, ...), those calls reached thunks that never checked this. They returned undefined where node throws a TypeError.

   perry/crates/perry-runtime/src/object/global_this/array_error.rs

   Lines 33 to 38 in ef6aee0

   	// `Function.prototype.call.call(undefined)` reaches here with the
   	// non-callable receiver as `this`; the value-call bridge would return
   	// `undefined` for it instead of the spec `TypeError`.
   	if unsafe { crate::object::fn_proto_receiver_not_callable(target) } {
   	crate::object::throw_fn_proto_not_callable("call");
   	}

2. Match node on the other regressions (diff). Function property delete, Intl subclass fields, the bind/call/apply alias fold, toLocaleString, util.inspect depth, Temporal constructor, the CJS exports.undefined wrapper, and the macOS hot-TLS asm.

3. Fix the gap harness and record iterret (diff). The harness strips the new GC banner and uses an expected file when Node crashes. Both snapshot changes come from CI's Linux reports: 2159 now passes, and iterret_generator_prototype is recorded against Generator objects own next/return/throw, so a patched %GeneratorPrototype% method is ignored #12148.

4. Give perry-ext-zlib's test runtime the stdlib features (diff). a139abd declared the dev-dependency without features, so the well_known guard test failed.

   perry/crates/perry-ext-zlib/Cargo.toml

   Lines 32 to 35 in ef6aee0

   	# Same feature set as the shipped runtime, as every perry-ext-* crate keeps it
   	# (#6303 `default`, #6314 `stdlib`); `ext_crates_bundle_a_full_featured_perry_runtime`
   	# in well_known.rs guards it.
   	perry-runtime = { workspace = true, features = ["default", "stdlib"] }

5. Keep three fixtures within their purpose (diff). 79dc03e chose miniz_oxide, whose bytes differ from node's, so the zlib fixtures check the round trip and the decoded CRC. The gc_11828 and cb-sqlite fixtures drop loops and churn that timed out on CI but test nothing extra.

6. Fix the gates that main's newest commits left red (diff). Each one also fails on main at 6521c0c.

   • warnings: accessor_entry_pair gets its only caller's cfg, and an unused import goes.
   • cargo-test: PERRY_B4_SABOTAGE changes emitted code, so it now keys the build cache.
   • Store-site, addr-class and string-payload gates: audit markers on new raw writes, is_handle_band in byte_word_address, and OwnedStringBytes in one test.
   • Inventories that still named deleted or rewritten code: shape census, thread-exit verdicts (new REGIONS map), registry floor, ephemeron seed rekey, native-handle ledger, and the rooting doc's workflow path.
   • weakmap-owned-index-validation.md is now a historical record for the knob gate. The run that it records used PERRY_GC_STRESS_SEED and PERRY_GC_STRESS_RATE, which no parser reads, so that run was not GC-stressed.

   perry/crates/perry-runtime/src/buffer/header.rs

   Lines 66 to 78 in ef6aee0

   	pub(crate) fn byte_word_address(bits: u64) -> Option<usize> {
   	let tagged = (bits & crate::value::TAG_MASK) == crate::value::POINTER_TAG;
   	if !tagged && (bits >> 48) != 0 {
   	return None;
   	}
   	let addr = (bits & crate::value::POINTER_MASK) as usize;
   	if crate::value::addr_class::is_handle_band(addr) {
   	return None;
   	}
   	if tagged {
   	return Some(addr);
   	}
   	raw_byte_word_is_owned(addr).then_some(addr)

7. Regenerate the call-effects tables (diff). macOS is generated locally. Linux and Windows come from CI's jobs on this branch, and CI's macOS output matches the local one.

🤖 Generated with Claude Code

https://claude.ai/code/session_01V5MSmMLJm6sMdSS5r64ds7

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

• Autopilot · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

---

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.}