fix(db): resolve ::-qualified call targets without breaking namespaced-name storage

[jrollin]

Problem

The heuristic edge resolver reduced target_name to a simple_name by splitting on . (method chains) and \ (PHP namespaces) but not :: (Rust/C++ path separator). So Rust path calls like Pool::new, Config::load, Vec::new never matched any bare-stored callee symbol and failed every one of the 6 resolution tiers — the largest driver of Rust being the lowest-scored language on make bench-resolution (~45%).

Fix: full-name-first, ::-reduced fallback through locality tiers only

A naive "just split on :: too" is not all-language safe — Ruby stores namespaced symbols under their full name (class Foo::Bar → symbol Foo::Bar, ruby.rs extract_constant_name keeps the whole scope_resolution), so reducing an Inherits target Baz::Quux → Quux would miss it and regress every Ruby namespaced inheritance/reference. It also lets a fully-qualified external call std::mem::swap reduce to swap and false-resolve onto a lone project symbol.

So the resolver now:

• full_name (split only on ./\, the prior behavior) is tried first — preserves qualified-name storage (Ruby Baz::Quux, PHP App\Auth\Foo).
• reduced_name (further past ::, Some only when it differs) is tried second, through locality tiers 1–4 only. Those are file/dir/scope-guarded, so a fallback can only match a symbol already reachable from the caller — safe.
• Global tier (5+6) uses full_name only: std::mem::swap stays unresolved rather than reducing to swap and fabricating a call edge.

Measurements (before → after, heuristic --no-lsp)

Lang	before	after	Δ
Rust	777 (45.47%)	808 (47.28%)	+31 (+1.81pp)
Ruby	388	388	0
Python / PHP / TS / Go / Java	—	—	0 (byte-identical)

Concrete: on the webapp_rs fixture, Config::load (10 call sites) resolved to nothing before (refs load → "No references found"); after, all 10 resolve. The gain is a correct +31, not a naive ::-split's inflated +68 — the 37-edge difference was exactly the false positives this design avoids. Index wall-clock unchanged (best-of-5 identical).

Regression-first tests

Each new test was verified against three resolver versions:

Test	plain main	naive ::-split (rejected)	this fix
Ruby namespaced inherits	✅	❌ regression	✅
std::mem::swap unresolved	✅	❌ false positive	✅
Rust gains (Pool::new, nested ::, dot-then-colon, View::render)	❌ no gain	✅	✅

Review

Ran an xhigh workflow-backed code review on the first (naive) attempt; it caught the Ruby regression and the external-call false-positive, both independently reproduced and now fixed with guarding tests. Also addressed the review's test-quality findings (mislabeled tier, redundant internal-column read, a false-green ambiguity test).

Checks

• cargo test --workspace (33 suites) ✅ — cartog-db 208 pass
• cargo test -p cartog-db --no-default-features ✅
• cargo fmt --check, cargo clippy --all-targets [--no-default-features] -- -D warnings ✅
• cargo test --doc -p cartog-db, cargo doc --no-deps -p cartog-db (warning-clean) ✅

No extractor / ground-truth / docs / site change: stored target_name is untouched, and this is an internal resolver-quality fix (no new command/flag/config/MCP-tool/language/count).

🤖 Generated with Claude Code

https://claude.ai/code/session_01AhoiweoMbZQ6FQgAcrfiWu

Summary by CodeRabbit

• Bug Fixes

  • Improved edge resolution for namespaced calls, making local matches more accurate when names include ::, . or path separators.
  • Reduced false positives for fully qualified external calls and ambiguous names, so unresolved references are handled more reliably.
  • Preserved correct precedence for nearby matches, such as same-file resolution over broader project-wide matches.

• Tests

  • Added regression coverage for namespaced resolution, ambiguous matches, and locality-based tie-breaking.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 25686ab7-19fc-4015-9db3-fb5d0f55f636

📥 Commits

Reviewing files that changed from the base of the PR and between caf64ab and 6828077.

📒 Files selected for processing (2)

• crates/cartog-db/src/store/resolution.rs
• crates/cartog-db/src/tests/resolution.rs

---

📝 Walkthrough

Walkthrough

The edge resolver in resolve_edge_batch was refactored to compute both a full_name and a reduced_name (split on ::) for target matching. Locality tiers 1-4 try full_name then fall back to reduced_name; global tiers 5-6 use full_name only. New regression tests validate these matching behaviors and provenance outcomes.

Changes

Edge Resolver Name Reduction

Layer / File(s)	Summary
Locality tier candidate resolution crates/cartog-db/src/store/resolution.rs	Replaces single simple_name with full_name/reduced_name computation and an ordered candidate loop for tiers 1-4, building an Option<(id, provenance)> instead of per-tier continue chains.
Global tier resolution restricted to full_name crates/cartog-db/src/store/resolution.rs	Updates tiers 5-6 to search project-wide using full_name only, avoiding reduced_name to prevent false positives from fully-qualified external calls.
Regression tests for name reduction and provenance crates/cartog-db/src/tests/resolution.rs	Adds a call_edge_target helper and tests covering same-file reduced-name resolution, global full-name resolution, ambiguous/false-positive cases, and mixed ./:: name forms.

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

Possibly related PRs

• jrollin/cartog#86: Also modifies resolve_edge_batch's tier-selection/update logic in the same file, extending resolution updates for provenance/resolution_source.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly matches the main change: resolving ::-qualified call targets while preserving namespaced-name storage.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
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 docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/resolver-colon-split-full-name-first

---

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.}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 88.38%. Comparing base (caf64ab) to head (6828077).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #148   +/-   ##
=======================================
  Coverage   88.38%   88.38%           
=======================================
  Files          96       96           
  Lines       32574    32573    -1     
=======================================
  Hits        28791    28791           
+ Misses       3783     3782    -1     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.