codegen: emit the receiver-unknown numeric index tiers once per dynamic site

[proggeramlug]

Rebased onto main after #8931 merged; one commit.

What

arr[key] on an erased-Array receiver with a dynamic key (packed[sparse[x]] in wolf-ecs SparseSet.has, a[b[i]] in general) lowers through the canonical-i32 split. Since the brand arm (#8890's v90), every integral key below 2³¹ — canonical double or INT32-boxed — is served in the canonical arm's aidx.claimed.other branch, which emits the receiver-unknown numeric tiers (inline typed-array read tav.*, dense Array-subclass arrlike.ic.*, complete dispatcher) once. The runtime-key arm (lower_claimable_array_string_key_get) still carried its own copy of that lattice from the v83 dynamic-key work (aidxkey.int → aidxkey.int.exact), which after v90 only ever ran for integral keys in [2³¹, 2³²). Its other caller passes static string/symbol keys, where the numeric path is dead.

This PR drops that copy: those keys take the complete js_array_get_index_or_string route (which handles every index). Same semantics; ~11 KB of IR less per dynamic site.

Why it matters

Diagnosing why SparseSet.has$pshape is still a bl inside addComponent$pshape (campaign item #4): an env-gated trace of the pre-statepoint inline admission showed the one-statement clone has$pshape$idx_u31 at 45,936 bytes of lowered IR against the 16 KiB budget (GUARDED_SPECIALIZATION_PREINLINE_MAX_IR_BYTES) — which is why the earlier 32 KiB experiment (v82b) could never fire. Byte split: arrlike.ic 13.2 KB, tav 9.4 KB, guarded arr 6.7 KB, aidx 5.7 KB, plen 5.1 KB — i.e. the dynamic-key site alone was ~30 KB because the two arms each emitted the same lattice. This is the first of the size reductions; the clone is still above budget afterwards (the .length IC lattice and the two guarded this.sparse[x] tiers are next), so no inlining change is claimed here — only smaller code at every such site.

Verification (local, per the campaign rule)

• The v83 test (any_typed_dynamic_key_takes_the_numeric_tiers_when_it_is_an_array_index) now asserts exactly one tav.get.brand and one arrlike.ic.family_token block definition per site, no aidxkey.int block, and the SSO-string and complete-route calls still present. index_get_claim_tests: 12/12.
• Crate suites (perry-codegen, perry-transform, perry-hir) and the lint gates (-D warnings, file size, GC store inventory, raw-handle debt, shape census, local-binding audit, addr-class audit) are running now; I'll confirm in a comment together with the Mac-mini paired screens (2 s + 50 ms windows, 11 pairs, both wolf-ecs benchmarks) and the traced clone sizes before/after.

https://claude.ai/code/session_019WVcWKmYsUBnnFB7nBgbBJ

Summary by CodeRabbit

• Performance

  • Reduced generated code for dynamic array access, improving code size in affected methods.
  • Eliminated redundant runtime dispatch for certain numeric array keys without changing behavior.

• Bug Fixes

  • Preserved correct handling for canonical numeric keys, string keys, and non-index keys.
  • Maintained complete fallback behavior for dynamic keys that are not recognized as array indexes.

• Tests

  • Updated coverage to verify streamlined dispatch and correct handling across supported key types.

[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: b509ba28-aa9e-43d0-91d5-050ca5ab07d7

📥 Commits

Reviewing files that changed from the base of the PR and between 0b78ea5 and 9e49235.

📒 Files selected for processing (3)

• changelog.d/0000-claimed-key-single-lattice.md
• crates/perry-codegen/src/expr/index_get.rs
• crates/perry-codegen/src/expr/index_get_claim_tests.rs

Included review availability: Your plan provides up to 8 included reviews per hour; 0 remain after this review.

---

📝 Walkthrough

Walkthrough

The compiler removes the duplicate receiver-unknown numeric dispatch for dynamic keys on erased arrays. Non-SSO keys now use js_array_get_index_or_string, while tests verify single-lattice emission and updated key classifications.

Changes

Claimed array key dispatch

Layer / File(s)	Summary
Update claimable-array lowering crates/perry-codegen/src/expr/index_get.rs	The non-SSO branch now uses the generic array index-or-string fallback. The merge no longer includes the removed inline numeric result.
Validate single-lattice emission crates/perry-codegen/src/expr/index_get_claim_tests.rs, changelog.d/0000-claimed-key-single-lattice.md	Tests verify updated classifications, one tav.* and arrlike.ic.* lattice per dynamic site, and the aidxkey.sso fallback route. The changelog records the change.

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

Merge Risk: ⚪ Minimal · up to 9e492

This localized codegen change removes duplicate numeric-tier emission while preserving the complete fallback behavior; no actionable merge-blocking risk remains at the current head beyond normal checks and review.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title is concise, specific, and accurately summarizes the main code generation change: emitting receiver-unknown numeric index tiers once per dynamic site.
Description check	✅ Passed	The description clearly explains the change, motivation, semantic impact, generated-IR reduction, affected code paths, and verification status. It does not use the template headings or include the che…
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 2 functions across 2 files. (1 skipped: 1 …
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.

Full details: Description check

Explanation

The description clearly explains the change, motivation, semantic impact, generated-IR reduction, affected code paths, and verification status. It does not use the template headings or include the checklist and related-issue section, but it is sufficiently complete and directly relevant.

Full details: Docstring Coverage

Explanation

Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 2 functions across 2 files. (1 skipped: 1 unsupported.)

✨ 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.}

[proggeramlug]

Local gates on 4ae1c462c: cargo test -p perry-codegen -p perry-transform -p perry-hir 2500 passed / 0 failed; RUSTFLAGS=-D warnings cargo check --workspace --all-targets (host-compatible scope) clean; file size OK; GC store-site inventory passed; raw-handle debt unchanged (967, no ceilings raised); shape census OK; local-binding audit OK; addr-class audit passed. Changelog fragment included.

#8931 has merged since this was opened — rebasing onto main next so only this PR's commit remains; Mac-mini paired screens and the before/after traced clone sizes to follow.

[proggeramlug]

Traced lowered-IR sizes (PERRY_PREINLINE_TRACE, a local diagnostic patch not in this PR) for the wolf-ecs add/remove tail, before → after this change; the pre-statepoint budget is 16 KiB, so nothing newly flips to admitted=true here — this is the size reduction only:

function	before	after	Δ
ECS.addComponent$pshape$generic	244,484	166,985	−31.7%
ECS.removeComponent$pshape$generic	244,087	166,576	−31.8%
ECS.updatePending$pshape	266,519	189,026	−29.1%
SparseSet.has$pshape$generic	102,314	63,572	−37.9%
SparseSet.remove$pshape$generic	68,312	42,935	−37.1%
SparseSet.has (u31 clone)	56,272	43,712	−22.3%
SparseSet.has$pshape (u31 clone)	45,936	33,583	−26.9%
ECS.destroyEntity$pshape$generic	52,510	40,177	−23.5%

21 other traced clones unchanged (no dynamic-key site). Machine-code effect and the Mac-mini screens follow once the build here finishes.

[proggeramlug]

Rebased onto main (9e492352d, one commit) after #8931 merged. Re-verified on the rebased head: RUSTFLAGS=-D warnings cargo check --workspace --all-targets clean; cargo test -p perry-codegen -p perry-transform -p perry-hir 2504 passed / 0 failed. Mac-mini screens are running; posting them when done.

[proggeramlug]

Mac-mini paired screens (11 alternating pairs, taskpolicy -t 0 -l 0), #8931 head (v95) → this PR (v98):

window	add/remove	entity cycle
2 s	0.4057 → 0.4059 (+0.03%, 4/11)	0.3365 → 0.3378 (+0.39%, 0/11)
50 ms	0.4050 → 0.4052 (+0.02%, 5/11)	0.3364 → 0.3377 (+0.34%, 1/11)

So: no time win — add/remove is noise-flat and the entity cycle is consistently ~0.4% slower (deltas 0.14…0.71 with one 1.53), most plausibly code layout after the hot has bodies shrank. The IR/machine-code reduction is real (table above) but by the campaign's retain rule (both benchmarks 11/11 better) this does not qualify on its own. Converting to draft: it stays open only because the follow-up single-statement pre-inline budget experiment (v99) needs the smaller clone to be admitted at all; if that experiment doesn't win, I'll close this.

[proggeramlug]

Follow-up experiment (v99: a 40 KiB pre-statepoint inline budget for one-statement bodies, which this PR's smaller has$pshape clone makes reachable) on the Mac, this PR (v98) → v99: add/remove −0.36% / −0.43% (11/11 in 2 s and 50 ms), entity cycle −0.22% / −0.30% (10/11 both). Net of the two changes vs main: add/remove ≈ −0.3%, entity ≈ +0.2% — nothing. The traces explain it: the hot add$pshape/remove$pshape clones already contain has$pshape$idx (index clones are always pre-statepoint-inlined), so what the budget admits is only the $generic fallback wrappers, and the dead lattices removed here cost compile size, not time.

Closing this PR: the change is correct and smaller, but it does not pay by the campaign's retain rule, and I'd rather not land layout-neutral code churn while three ECS PRs are in flight. The branch stays (perf/claimed-key-single-lattice, 9e492352d) in case a later change needs the freed budget.