fix(codegen): coerce declared-only addition results

[proggeramlug]

Summary

Fix the raw-f64 guard-failure path for typed receiver methods when a declared-number + result feeds another arithmetic operator.

After the field guard falls back to a $pshape clone, this.x + this.y can correctly produce a boxed string when a runtime field violates its declared numeric type. Two paths then lost that fact:

• a TypeScript-inferred numeric local such as const sum = this.x + this.y was not marked declared-only because the tracker only covered Any -> Number refinements;
• a direct compound operand such as (this.x + this.y) * scale was excluded because residual coercion only consulted declared-only proof for LocalGet.

This change propagates declared-only proof into every numeric local whose initializer carries it, and consults the same proof for every arithmetic operand. Proven raw-f64 tiers still answer false and remain coercion-free.

The existing clone-composition test now asserts the semantic order dynamic + -> ToNumber -> multiply and that the typed raw-f64 clone retains no dynamic add or number coercion. A parsed TypeScript parity regression covers both intermediate-local and direct-expression forms through the actual $pshape fallback.

Reproduction

Both compiler arms linked generated programs against the same runtime:

pre-fix Perry: 12 string 12 string
fixed Perry:   36 number 36 number
Node:          36 number 36 number

Validation

• cargo test -p perry-codegen --test native_proof_regressions -- --test-threads=1 — 262 passed
• focused typed_f64_receiver_method_clone_raw_loads_after_composed_guards — passed
• release compiler build — passed
• two-compiler / one-runtime executable A/B — fixed output matches Node
• python3 scripts/check_test_registration.py — passed
• cargo fmt --all -- --check — passed
• bash scripts/check_file_size.sh — passed
• cargo test -p perry-codegen — 877 library tests passed, then stopped at the documented pre-existing large_local_array_push_inbounds_store_emits_precise_slot_barrier failure

Closes #7506.

Summary by CodeRabbit

• Bug Fixes

  • Improved numeric coercion for calculations involving declared numeric fields, compound expressions, and values that change type at runtime.
  • Fixed arithmetic when numeric fields are replaced with strings, ensuring results are coerced correctly before subsequent calculations.
  • Preserved optimized behavior for values already proven to be valid numbers.

• Tests

  • Added regression coverage for regular and inline method calls, runtime type changes, and fallback arithmetic 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: 5d826d16-5044-4cc1-ad4e-85030980e2f3

📥 Commits

Reviewing files that changed from the base of the PR and between 2c5a4ef and 6903f0b.

📒 Files selected for processing (1)

• crates/perry-codegen/tests/native_proof_regressions.rs

🚧 Files skipped from review as they are similar to previous changes (1)

• crates/perry-codegen/tests/native_proof_regressions.rs

---

📝 Walkthrough

Walkthrough

The compiler extends declared-only numeric proof tracking to numeric locals and compound expressions. Residual coercion now preserves JavaScript semantics in typed-receiver fallbacks. Regression tests verify coercion and fallback ordering.

Changes

Declared-only numeric coercion

Layer / File(s)	Summary
Declared-only proof tracking crates/perry-codegen/src/stmt/let_stmt.rs, crates/perry-codegen/src/type_analysis/pod.rs, crates/perry-codegen/src/expr/mod.rs	Numeric locals from declared-only initializers are tracked for refined and directly numeric types. Documentation comments describe compound expressions and boxed-value propagation.
Residual coercion and regression coverage crates/perry-codegen/src/expr/binary.rs, crates/perry-codegen/tests/native_proof_regressions.rs, test-files/test_gap_7506_pshape_add_result_coercion.ts, changelog.d/7851-pshape-add-result-coercion.md	Residual coercion now uses numeric_proof_is_declared_only. Regression checks validate dynamic addition, js_number_coerce, multiplication ordering, poisoned numeric fields, and typed receiver fallback behavior. The changelog records the coercion fix.

Estimated code review effort: 3 (Moderate) | ~20 minutes

Sequence Diagram(s)

sequenceDiagram
  participant NumericLocal
  participant operand_needs_residual_coerce
  participant js_number_coerce
  participant ReceiverFallback
  NumericLocal->>operand_needs_residual_coerce: provide declared-only numeric expression
  operand_needs_residual_coerce->>js_number_coerce: coerce residual numeric value
  ReceiverFallback->>js_number_coerce: coerce dynamic addition result
  js_number_coerce-->>ReceiverFallback: return numeric value for multiplication

Loading

Possibly related PRs

• PerryTS/perry#7831: Extends the same declared-only numeric proof and residual coercion paths.
• PerryTS/perry#7842: Also updates numeric operator lowering and declaration-only proof handling.
• PerryTS/perry#7669: Updates numeric-proof and arithmetic lowering in the same codegen area with a different proof source.

Suggested labels: bug

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and concisely describes the main code generation fix.
Description check	✅ Passed	The description explains the problem, solution, affected behavior, linked issue, and validation results.
Linked Issues check	✅ Passed	The changes address issue #7506 by validating composed guard behavior and preserving a correct dynamic fallback for guard failures.
Out of Scope Changes check	✅ Passed	The code, documentation, regression tests, test fixture, and changelog entry all support the stated issue and PR objectives.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

✨ 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 fix/7506-pshape-fallback-coercion

---

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]

Actionable comments posted: 1

🤖 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.

Inline comments:
In `@crates/perry-codegen/tests/native_proof_regressions.rs`:
- Around line 11746-11751: Update the regression assertions around the cond_br
consuming js_typed_feedback_class_field_get_guard so the typed clone is required
only in the guard-success successor and explicitly rejected in the failure
successor. Replace the unconditional call-pattern check that currently accepts
any i64 receiver with successor-specific validation, ensuring $typed_f64_recv
cannot appear on the failed field-guard path while preserving the existing
coercion checks there.

🪄 Autofix

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 343d855e-de06-466e-91ec-4242610d1557

📥 Commits

Reviewing files that changed from the base of the PR and between 079e646 and d091765.

📒 Files selected for processing (6)

• crates/perry-codegen/src/expr/binary.rs
• crates/perry-codegen/src/expr/mod.rs
• crates/perry-codegen/src/stmt/let_stmt.rs
• crates/perry-codegen/src/type_analysis/pod.rs
• crates/perry-codegen/tests/native_proof_regressions.rs
• test-files/test_gap_7506_pshape_add_result_coercion.ts