perf(codegen): lower proven booleans into native numeric ops

[proggeramlug]

Summary

• convert guard-proven Boolean i1 literals and locals directly to f64 with uitofp for arithmetic consumers
• lower proven Boolean/number relational comparisons to native fcmp instead of js_rel_*
• keep annotation-only and invalidated Boolean facts on dynamic coercion/dispatch, preserving as any semantics
• add IR-census coverage plus Boolean arithmetic parity coverage

This is a focused Lever E slice that advances #5497; it does not close the broader tracker.

Validation

• cargo test -p perry-codegen boolean_number_tests --lib
• cargo test -p perry-codegen --lib (967 passed)
• ./scripts/test_affected_crates.sh --base origin/main
• Node 26.5.1 parity: test_bool_arithmetic (1/1 passed)
• python3 scripts/check_test_registration.py
• ./scripts/pre-tag-check.sh --quick (all relevant checks passed; the run reports the existing GC inventory ratchet at unchanged crates/perry-codegen/src/expr/property_set.rs:1457)

No version files were changed. The required PR-numbered changelog fragment is included.

Summary by CodeRabbit

• New Features

  • Improved arithmetic and comparison handling for statically proven Boolean values.
  • Boolean values can now convert directly to numbers for supported numeric operations.
  • Added optimized Boolean-to-number relational comparisons.

• Bug Fixes

  • Preserved dynamic coercion and string concatenation when Boolean values are invalidated or cannot be safely proven numeric.

• Tests

  • Added coverage for Boolean arithmetic, comparisons, nested expressions, and invalidated Boolean annotations.

• Documentation

  • Updated the type-lowering checklist and changelog with the new behavior.

[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: ef7f7ec3-a05b-4a23-bc45-62c9637b8d7e

📥 Commits

Reviewing files that changed from the base of the PR and between 8b40f26 and 490958a.

📒 Files selected for processing (8)

• TYPE_LOWERING.md
• changelog.d/8031-boolean-native-numeric-ops.md
• crates/perry-codegen/src/expr/binary.rs
• crates/perry-codegen/src/expr/boolean_number_tests.rs
• crates/perry-codegen/src/expr/mod.rs
• crates/perry-codegen/src/stmt/let_stmt.rs
• crates/perry-codegen/src/type_analysis/numeric.rs
• test-files/test_bool_arithmetic.ts

---

📝 Walkthrough

Walkthrough

Proven native Booleans now convert directly to f64 for arithmetic and Boolean/number comparisons. Declared-only or invalidated Boolean and numeric proofs retain dynamic coercion and dispatch. New IR-census and end-to-end tests cover both paths.

Changes

Boolean numeric lowering

Layer / File(s)	Summary
Proven Boolean conversion and comparison crates/perry-codegen/src/expr/mod.rs	Proven Boolean literals, locals, and compiler-private async controls convert from i1 to f64 with uitofp. Boolean/number comparisons use native ordered fcmp when operand proofs are valid.
Arithmetic dispatch integration crates/perry-codegen/src/type_analysis/numeric.rs, crates/perry-codegen/src/stmt/let_stmt.rs, crates/perry-codegen/src/expr/binary.rs	Canonical raw-f64 analysis excludes declared-only binary proofs. Numeric locals track non-numeric initializers. Arithmetic lowering uses native Boolean conversion only for proven-safe operands.
Lowering validation and acceptance evidence crates/perry-codegen/src/expr/boolean_number_tests.rs, test-files/test_bool_arithmetic.ts, TYPE_LOWERING.md, changelog.d/8031-boolean-native-numeric-ops.md	Tests and documentation cover native Boolean arithmetic, relational comparisons, nested declared-only expressions, and dynamic fallback after string assignment.

Estimated code review effort: 3 (Moderate) | ~25 minutes

Mergeability Score: ⚪ Minimal · up to 49095

The PR makes a focused boolean numeric-lowering change with passing validation; no actionable merge-blocking risk remains beyond normal checks and review.

Sequence Diagram(s)

sequenceDiagram
  participant BooleanExpression
  participant ExprLowering
  participant BinaryLowering
  participant LLVM
  BooleanExpression->>ExprLowering: provide proven i1 value
  ExprLowering->>LLVM: emit uitofp i1 to f64
  BinaryLowering->>LLVM: emit native arithmetic or ordered fcmp
  BooleanExpression->>BinaryLowering: provide invalidated or declared-only value
  BinaryLowering->>LLVM: retain dynamic coercion and dispatch

Loading

Possibly related PRs

• PerryTS/perry#7831: Extends its declared-only numeric tracking and lowering safeguards.
• PerryTS/perry#6860: Uses related proven-type fast paths in binary numeric code generation.
• PerryTS/perry#7839: Addresses related + dispatch guards for erased TypeScript type declarations.

Suggested reviewers: thehypnoo

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and concisely describes the main codegen change: lowering proven booleans into native numeric operations.
Description check	✅ Passed	The description explains the changes, scope, related issue, and validation steps, but it omits explicit template headings for Changes, Related issue, and Checklist.
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

---

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]

Blocking on exact head 7a1f654. I reproduced a semantic regression with a focused IR test: let flag = true; let liar: number = "4" as any; flag < (liar + 1). JavaScript evaluates this as true because the RHS is "41". This head emits a raw fadd on the NaN-boxed string followed by fcmp olt, producing false, and omits js_rel_lt. The new relational gate accepts the nested Add because expr_produces_canonical_raw_f64 blanket-admits Binary via type-level is_numeric_expr, even though a declared-only Add tree can carry a boxed/string result. Please make the Boolean gate reject declared-only/noncanonical nested binary trees and add this regression. This is independent proof from emitted IR, not a CI result. No version bump needed.

[proggeramlug]

Addressed in 490958a.

The fix makes the existing proof boundary explicit in both places involved:

• a numeric local initialized by a non-numeric value (for example const liar: number = "4" as any) now carries the declared-only hazard;
• expr_produces_canonical_raw_f64 rejects binary trees carrying that hazard, so the Boolean relational fast path cannot admit the nested +.

I added the exact flag < (liar + 1) IR regression. It now requires both js_dynamic_string_or_number_add and js_rel_lt, while the positive flag < 2 case still requires native uitofp + fcmp.

Validated with all 968 perry-codegen library tests, the affected-crate sweep, and the optimized test_bool_arithmetic parity fixture (1/1 pass). No version bump.

[proggeramlug]

Re-audited exact head 490958a against the previously demonstrated nested declared-only numeric regression. The corrected IR retains both js_dynamic_string_or_number_add and js_rel_lt for a Number-annotated local holding a string, while the positive proven-boolean case retains native fcmp. The new declared-only seed and canonical-f64 rejection are aligned, the focused boolean-number tests pass, and the full perry-codegen library suite passes (968/968). No blocking issue found.