refactor(transform/hir): de-duplicate HIR-walking helpers (#5293)

[proggeramlug]

Closes #5293.

An external audit flagged several HIR/IR-walking helpers copy-pasted across files. Each copy must be hand-updated whenever a new Expr/HIR node variant is added; a missed copy is a silent miscompile, not a build error — the same bug-class behind #5143 (FuncId-collision from a scan missing class field-init closures).

What changed

1. compute_max_local_id / compute_max_func_id (perry-transform)
Deleted the copies in finally_inline, async_to_generator (including compute_max_func_id_module), state_desugar, and unroll, plus their private scan_* helpers. All callers now use the canonical, already-pub, exhaustive-walk_expr_children-backed implementations in generator::id_scan.

Before routing, the canonical scanners were made a true superset of every copy so no caller loses coverage:

• scan switch case.test expressions (some copies did, the canonical didn't),
• scan module-global initializers (async_to_generator's copy did, the canonical didn't).

A higher max id is always safe — fresh ids just start higher; a missed id is what collides.

2. class_computed_member_registration_expr (perry-hir)
Dropped the 3 byte-identical copies in lower/{lower_expr,module_decl,stmt}.rs and re-exported the canonical lower_decl::class_computed one through lower_decl::mod (the lower/* files already pull it in via use super::*).

3. collect_assigned_locals_expr (perry-hir)
Replaced the ~1120-line hand-rolled per-variant descent — which ended in a _ => {} catch-all that silently skipped any newly-added variant nesting a LocalSet/Update — with the exhaustive walk_expr_children for child descent. Only the 7 assignment-recording arms (LocalSet, Update, array push/splice/copyWithin/reverse, SetAdd) and the intentional don't-recurse-into-closures arm remain custom. Net: same recorded assignments on every currently-handled variant, plus coverage for variants the catch-all used to drop.

The optional codegen for_each_body_in_module extraction (item 4 in the issue) is not included here — left for a follow-up to keep this diff a pure mechanical dedup.

Validation

Pure refactor, no behavior change. Net −2176 lines.

• perry-hir lib (158) and perry-transform lib (34) tests pass in release (includes the id_scan unit tests and the lower: deeply-nested expressions (a+b+c+…, o.a.a.…) overflow the stack → fatal abort instead of a diagnostic #5259 deep-chain guard tests).
• An end-to-end smoke program exercising switch case-tests, module-global closures, async/await, generators, try/finally, computed class members, and unrolled loops compiles and matches node --experimental-strip-types byte-for-byte.

🤖 Generated with Claude Code

Summary by CodeRabbit

• Bug Fixes
  • Reduced the risk of LocalId/FuncId allocation collisions by expanding max-id scanning to cover additional expression locations (including global initializers and switch case tests) and by using shared canonical ID computation helpers across transformation passes.
• Refactor
  • Consolidated duplicate “max id”/scan logic across multiple transformation passes.
  • Streamlined assignment-tracking and reorganized class computed-member lowering helpers for clearer, more reliable analysis.

[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: fe6a4922-15ee-4a1c-834d-2baf36a932ca

📥 Commits

Reviewing files that changed from the base of the PR and between 6a76aa0 and 1c7ba04.

📒 Files selected for processing (1)

• crates/perry-hir/src/analysis.rs

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

• crates/perry-hir/src/analysis.rs

---

📝 Walkthrough

Walkthrough

Two independent deduplication refactors from issue #5293: (1) in perry-hir, three copy-pasted class_computed_member_registration_expr helpers are removed from lower/ sub-modules and routed through a single pub(crate) re-export, and collect_assigned_locals_expr is rewritten from ~1000 lines of exhaustive match to a slim assignment-focused match plus walk_expr_children; (2) in perry-transform, duplicated compute_max_local_id/compute_max_func_id implementations are deleted from four passes and replaced with imports of the canonical generator/id_scan helpers, which are also extended to scan global.init and Stmt::Switch case-test expressions.

Changes

HIR: class_computed helper consolidation and analysis refactor

Layer / File(s)	Summary
Consolidate class_computed_member_registration_expr to single canonical source crates/perry-hir/src/lower_decl/mod.rs, crates/perry-hir/src/lower/lower_expr.rs, crates/perry-hir/src/lower/module_decl.rs, crates/perry-hir/src/lower/stmt.rs	Adds pub(crate) use class_computed::class_computed_member_registration_expr in lower_decl/mod.rs and removes the three identical private copies from lower_expr.rs, module_decl.rs, and stmt.rs.
Refactor collect_assigned_locals_expr to use walk_expr_children crates/perry-hir/src/analysis.rs	Replaces ~1000-line exhaustive variant enumeration with a focused match on assignment-bearing expressions (LocalSet, Update, array-rebinding mutators, SetAdd, Closure guard), delegating all other child descent to walk_expr_children.

Transform: canonical id_scan extensions and duplicate removal

Layer / File(s)	Summary
Extend canonical id_scan to cover global.init and Switch case-test crates/perry-transform/src/generator/id_scan.rs	Adds global.init expression scanning and per-case test-expression scanning to both compute_max_local_id and compute_max_func_id, closing coverage gaps that allowed ID collisions when closures appeared in those positions.
Remove duplicated max-ID helpers from transform passes crates/perry-transform/src/async_to_generator.rs, crates/perry-transform/src/finally_inline.rs, crates/perry-transform/src/state_desugar.rs, crates/perry-transform/src/unroll/mod.rs	Deletes ~1000 lines of duplicated compute_max_local_id/compute_max_func_id and scan_* helpers from four transform passes, replacing each with an import of the canonical crate::generator helpers.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

Possibly related PRs

• PerryTS/perry#5186: Modifies compute_max_func_id / async_to_generator to add class-field initializer locations, directly overlapping with the ID-scan coverage extended in this PR.

Poem

🐇 Five scanners were five too many,
Each copied with care, not a penny —
Now one rules the crate,
And walkers delegate,
No silent miss-match, not any! ✨

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title accurately describes the main change: de-duplicating HIR-walking helpers as a refactoring effort targeting issue #5293.
Description check	✅ Passed	The PR description comprehensively covers the changes across three main components, includes the issue reference, validation approach, and follows repository conventions.
Linked Issues check	✅ Passed	All changes directly address the objectives from issue #5293: consolidates compute_max_local_id/compute_max_func_id copies [#5293], eliminates class_computed_member_registration_expr duplicates [#5293], and refactors collect_assigned_locals_expr [#5293].
Out of Scope Changes check	✅ Passed	All changes are scoped to the declared objectives: helper consolidation, canonical scanner enhancement (switch case.test, module-global initializers), and walk_expr_children delegation. Optional codegen extraction deferred as noted.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ 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 worktree-refactor-hir-walkers-5293

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}