feat: parser refactoring, Jedi call resolution, and performance optimizations

Major improvements to code-review-graph spanning parser architecture,
call graph accuracy, and build performance.

Parser refactoring:
- Extract 16 per-language handler modules into code_review_graph/lang/
  using a strategy pattern, replacing monolithic conditionals in parser.py
- Thread-safe parser caches with double-check locking

Call graph enrichment:
- Jedi-based Python method call resolution at build time (jedi_resolver.py)
- Pre-scan filtering by project function names (36s to 3s on large repos)
- Typed variable call enrichment (Python, JS/TS, Kotlin/Java)
- Star import resolution, namespace imports, CommonJS require()
- Angular template parsing, JSX handler tracking
- Module-level import tracking and module-qualified call resolution
- Function/class references passed as call arguments

PreToolUse search enrichment:
- New enrich.py module and code-review-graph enrich CLI command
- Injects graph context (callers, flows, community, tests) into agent
  search results passively via hook

Dead code false positive reduction:
- Framework decorators recognized as entry points
- CDK construct methods, abstract overrides excluded
- E2e test directories excluded from dead code detection

Performance:
- Community detection: 48.6s to 2.3s (21x speedup) via bulk node
  loading and adjacency-indexed cohesion computation
- Jedi enrichment: 36s to 3s (12x) via pre-scan filtering
- Batch file storage (50-file transactions)
- Batch risk_index (2 GROUP BY queries replace per-node loops)

Other:
- Weighted flow risk scoring by criticality
- Transitive TESTED_BY lookup for tests_for and risk scoring
- DB schema v8: composite edge index (v7 reserved by PR tirth8205#127)
- --quiet and --json CLI flags
- Search query deduplication, test function deprioritization
- New [enrichment] optional dependency group for Jedi
- 829+ tests across 26 test files (up from 615)

Evaluated against Gadgetbridge (41k nodes, 280k edges): 8/10 PASS,
call resolution rate improved from 28% to 39.6%.

Author: gzenz

CHANGELOG.md

@@ -55,14 +55,50 @@ Hotfix on top of 2.2.3 for two bugs surfaced by a full first-time-user smoke tes
 
 ### Added
 - **Codex platform install support** (PR #177): `code-review-graph install --platform codex` appends a `mcp_servers.code-review-graph` section to `~/.codex/config.toml` without overwriting existing Codex settings.
-- **Luau language support** (PR #165, closes #153): Roblox Luau (`.luau`) parsing — functions, classes, local functions, requires, tests.
+- **Luau language support** (PR #165, closes #153): Roblox Luau (`.luau`) parsing -- functions, classes, local functions, requires, tests.
 - **REFERENCES edge type** (PR #217): New edge kind for symbol references that aren't direct calls (map/dispatch lookups, string-keyed handlers), including Python and TypeScript patterns.
 - **`recurse_submodules` build option** (PR #215): Build/update can now optionally recurse into git submodules.
 - **`.gitignore` default for `.code-review-graph/`** (PR #185): Fresh installs automatically add the SQLite DB directory to `.gitignore` so the database isn't accidentally committed.
 - **Clearer gitignore docs** (PR #171, closes #157): Documentation now spells out that `code-review-graph` already respects `.gitignore` via `git ls-files`.
+- **Parser refactoring**: Extracted 16 per-language handler modules into `code_review_graph/lang/` package using a strategy pattern, replacing monolithic conditionals in `parser.py`
+- **Jedi-based call resolution**: New `jedi_resolver.py` module resolves Python method calls at build time via Jedi static analysis, with pre-scan filtering by project function names (36s to 3s on large repos)
+- **PreToolUse search enrichment**: New `enrich.py` module and `code-review-graph enrich` CLI command inject graph context (callers, callees, flows, community, tests) into agent search results passively
+- **Typed variable call enrichment**: Track constructor-based type inference and instance method calls for Python, JS/TS, and Kotlin/Java
+- **Star import resolution**: Resolve `from module import *` by scanning target module's exported names
+- **Namespace imports**: Track `import * as X from 'module'` and CommonJS `require()` patterns
+- **Angular template parsing**: Extract call targets from Angular component templates
+- **JSX handler tracking**: Detect function/class references passed as JSX event handler props
+- **Framework decorator recognition**: Identify entry points decorated with `@app.route`, `@router.get`, `@cli.command`, etc., reducing dead code false positives
+- **Module-level import tracking**: Track module-qualified call resolution (`module.function()`)
+- **Thread safety**: Double-check locking on parser caches (`_type_sets`, `_get_parser`, `_resolve_module_to_file`, `_get_exported_names`)
+- **Batch file storage**: `store_file_batch()` groups file insertions into 50-file transactions for faster builds
+- **Bulk node loading**: `get_all_nodes()` replaces per-file SQL queries for community detection
+- **Adjacency-indexed cohesion**: Community cohesion computed in O(community-edges) instead of O(all-edges), yielding 21x speedup (48.6s to 2.3s on 41k-node repos)
+- **Phase timing instrumentation**: `time.perf_counter()` timing at INFO level for all build phases
+- **Batch risk_index**: 2 GROUP BY queries replace per-node COUNT loops in risk scoring
+- **Weighted flow risk scoring**: Risk scores weighted by flow criticality instead of flat edge counts
+- **Transitive TESTED_BY lookup**: `tests_for` and risk scoring follow transitive test relationships
+- **DB schema v8**: Composite edge index for upsert performance (v7 reserved by upstream PR #127)
+- **`--quiet` and `--json` CLI flags**: Machine-readable output for `build`, `update`, `status`
+- **829+ tests** across 26 test files (up from 615), including `test_pain_points.py` (1,587 lines TDD suite), `test_hardened.py` (467 lines), `test_enrich.py` (237 lines)
+- **14 new test fixtures**: Kotlin, Java, TypeScript, JSX, Python resolution scenarios
 
 ### Changed
-- Community detection is now bounded — large repos complete in reasonable time instead of hanging indefinitely.
+- Community detection is now bounded -- large repos complete in reasonable time instead of hanging indefinitely.
+- New `[enrichment]` optional dependency group for Jedi-based Python call resolution
+- Leiden community detection scales resolution parameter with graph size
+- Adaptive directory-based fallback for community detection when Leiden produces poor clusters
+- Search query deduplication and test function deprioritization
+
+### Fixed
+- **Dead code false positives**: Decorators, CDK construct methods, abstract overrides, and overriding methods with called parents no longer flagged as dead
+- **E2e test exclusion**: Playwright/Cypress e2e test directories excluded from dead code detection
+- **Unique-name plausible caller optimization**: Faster dead code analysis via pre-filtered candidate sets
+- **Store cache liveness check**: Cached SQLite connections verified as alive before reuse
+
+### Performance
+- **Community detection**: 48.6s to 2.3s (21x) on Gadgetbridge (41k nodes, 280k edges)
+- **Jedi enrichment**: 36s to 3s (12x) via pre-scan filtering by project function names
 
 ## [2.2.2] - 2026-04-08
 

CLAUDE.md

@@ -46,7 +46,7 @@ When using code-review-graph MCP tools, follow these rules:
 
 ```bash
 # Development
-uv run pytest tests/ --tb=short -q          # Run tests (572 tests)
+uv run pytest tests/ --tb=short -q          # Run tests (609 tests)
 uv run ruff check code_review_graph/        # Lint
 uv run mypy code_review_graph/ --ignore-missing-imports --no-strict-optional
 

README.md

@@ -230,6 +230,7 @@ code-review-graph watch            # Auto-update on file changes
 code-review-graph visualize        # Generate interactive HTML graph
 code-review-graph wiki             # Generate markdown wiki from communities
 code-review-graph detect-changes   # Risk-scored change impact analysis
+code-review-graph enrich           # Enrich search results with graph context
 code-review-graph register <path>  # Register repo in multi-repo registry
 code-review-graph unregister <id>  # Remove repo from registry
 code-review-graph repos            # List registered repositories
@@ -296,6 +297,7 @@ Optional dependency groups:
 pip install code-review-graph[embeddings]          # Local vector embeddings (sentence-transformers)
 pip install code-review-graph[google-embeddings]   # Google Gemini embeddings
 pip install code-review-graph[communities]         # Community detection (igraph)
+pip install code-review-graph[enrichment]          # Jedi-based Python call resolution
 pip install code-review-graph[eval]                # Evaluation benchmarks (matplotlib)
 pip install code-review-graph[wiki]                # Wiki generation with LLM summaries (ollama)
 pip install code-review-graph[all]                 # All optional dependencies
@@ -319,7 +321,7 @@ pytest
 <summary><strong>Adding a new language</strong></summary>
 <br>
 
-Edit `code_review_graph/parser.py` and add your extension to `EXTENSION_TO_LANGUAGE` along with node type mappings in `_CLASS_TYPES`, `_FUNCTION_TYPES`, `_IMPORT_TYPES`, and `_CALL_TYPES`. Include a test fixture and open a PR.
+Edit the appropriate language handler in `code_review_graph/lang/` (e.g., `_python.py`, `_kotlin.py`) or create a new one following `_base.py`. Add your extension to `EXTENSION_TO_LANGUAGE` in `parser.py`, include a test fixture, and open a PR.
 
 </details>
 

code-review-graph-vscode/src/backend/cli.ts

@@ -53,22 +53,14 @@ export class CliWrapper {
     /**
      * Build (or fully rebuild) the graph database for a workspace.
      */
-    async buildGraph(
-        workspaceRoot: string,
-        options?: { fullRebuild?: boolean },
-    ): Promise<CliResult> {
-        const args = ['build'];
-        if (options?.fullRebuild) {
-            args.push('--full');
-        }
-
+    async buildGraph(workspaceRoot: string): Promise<CliResult> {
         return vscode.window.withProgress(
             {
                 location: vscode.ProgressLocation.Notification,
                 title: 'Code Review Graph: Building graph\u2026',
                 cancellable: false,
             },
-            () => this.exec(args, workspaceRoot),
+            () => this.exec(['build'], workspaceRoot),
         );
     }
 

code-review-graph-vscode/src/backend/sqlite.ts

@@ -212,7 +212,7 @@ export class SqliteReader {
         if (row) {
           const version = parseInt(row.value, 10);
           // Must match LATEST_VERSION in code_review_graph/migrations.py
-          const SUPPORTED_SCHEMA_VERSION = 6;
+          const SUPPORTED_SCHEMA_VERSION = 8;
           if (!isNaN(version) && version > SUPPORTED_SCHEMA_VERSION) {
             return `Database was created with a newer version (schema v${version}). Update the extension.`;
           }

code_review_graph/changes.py

@@ -154,15 +154,19 @@ def compute_risk_score(store: GraphStore, node: GraphNode) -> float:
     Scoring factors:
       - Flow participation: 0.05 per flow membership, capped at 0.25
       - Community crossing: 0.05 per caller from a different community, capped at 0.15
-      - Test coverage: 0.30 if no TESTED_BY edges, 0.05 if tested
+      - Test coverage: 0.30 (untested) scaling down to 0.05 (5+ TESTED_BY edges)
       - Security sensitivity: 0.20 if name matches security keywords
       - Caller count: callers / 20, capped at 0.10
     """
     score = 0.0
 
-    # --- Flow participation (cap 0.25) ---
-    flow_count = store.count_flow_memberships(node.id)
-    score += min(flow_count * 0.05, 0.25)
+    # --- Flow participation (cap 0.25), weighted by criticality ---
+    flow_criticalities = store.get_flow_criticalities_for_node(node.id)
+    if flow_criticalities:
+        score += min(sum(flow_criticalities), 0.25)
+    else:
+        flow_count = store.count_flow_memberships(node.id)
+        score += min(flow_count * 0.05, 0.25)
 
     # --- Community crossing (cap 0.15) ---
     callers = store.get_edges_by_target(node.qualified_name)
@@ -179,10 +183,10 @@ def compute_risk_score(store: GraphStore, node: GraphNode) -> float:
                 cross_community += 1
     score += min(cross_community * 0.05, 0.15)
 
-    # --- Test coverage ---
-    tested_edges = store.get_edges_by_target(node.qualified_name)
-    has_test = any(e.kind == "TESTED_BY" for e in tested_edges)
-    score += 0.05 if has_test else 0.30
+    # --- Test coverage (direct + transitive) ---
+    transitive_tests = store.get_transitive_tests(node.qualified_name)
+    test_count = len(transitive_tests)
+    score += 0.30 - (min(test_count / 5.0, 1.0) * 0.25)
 
     # --- Security sensitivity ---
     name_lower = node.name.lower()