[squid-protocol]

Summary

Extends #1264's per-language class_start named-entity extraction to rust, per epic #1295. No language_standards.py regex change needed -- two ground-truth-side fixes in tests/tools/tree_sitter_accuracy_audit.py:

1. NODE_MAPS["rust"]["class_node_types"] was missing union_item (tree-sitter-rust's node type for union Foo {...}, distinct from struct_item/enum_item) -- a real, named class-like entity rust's own class_start regex already matches (its comment covers struct/enum/union/trait). Confirmed via wasmtime's FRegUnion/VRegUnion/XRegUnion, plain top-level unions with no macro involvement. Adding it: real_classes/found_classes 242→245.
2. The remaining extra_classes (11→25 after the union fix, using tests/tools/class_start_diff.py --lang rust) all traced to struct/enum/trait declarations sitting inside a macro invocation or macro_rules!/proc-macro-template body -- cfg_not_rt! { pub(crate) struct JoinHandle<R> {...} }, pin_project! { pub enum MaybeDone<Fut: Future> {...} }, ast_struct! { pub struct Variant {...} }, serde's $typaram-templated visitor structs. Tree-sitter treats macro bodies as opaque token trees it can't parse into real declaration nodes, so real_classes can never see these, while GitGalaxy's regex (raw text, no macro-awareness) matches them fine. This is the same blind spot already documented for rust/swift func_start extra_functions rose after #1302's prism.py fix -- newly exposed, not caused #1311/rust func_start misses bodyless multi-line trait method signatures (deserialize_* family) #1319's extra_functions on rust -- now confirmed to apply identically to extra_classes. Not fixable without real macro expansion; documented as recurring-cause class 8 in the epic's checklist doc.

found_classes 165→245 (was low because rust has no class keyword, so the legacy fallback regex -- which only matches class|struct|interface|trait|enum -- structurally couldn't see most rust declarations). Baseline hand-blessed to the measured extra_classes=25, same precedent as csharp (#1264) and zig (#1367): a confirmed, understood, non-bug gap, not an unexplained regression.

Verification

• python tests/tools/tree_sitter_accuracy_audit.py --lang rust --ci -- clean against hand-blessed baseline
• python tests/tools/crucible_check.py -- both modes (full_precision, zero_dependency) PASS
• python tests/tools/audit_check.py --regenerate -- ruff findings were pure line-shifts, baseline regenerated
• python -m pytest tests/core_engine/ tests/extraction/ -- 6585 passed, 2 skipped, 11 xfailed, 3 xpassed

Test plan

• tree_sitter_accuracy_audit.py --lang rust --ci clean
• crucible_check.py both modes clean
• audit_check.py clean
• Full core_engine/extraction suite passes

🤖 Generated with Claude Code

[github-actions]

🐦‍⬛ Muninn Security Scan

✅ No security issues found.

🐦‍⬛ Powered by Muninn · Skald Lab