fix: Increase explore traversal depth and support qualified symbol lookups

colbymchenry · claude · colbymchenry · commit b986b78fa9c0 · 2026-04-03T19:56:58.000-05:00
Two fixes discovered while benchmarking Swift (Alamofire):

1. codegraph_explore traversalDepth 2→3: Deep call chains (e.g., Alamofire's
   9-step Session.request()→URLSession flow) couldn't be followed in a single
   explore call, forcing agents to fall back to file reads.

2. findSymbol/findAllSymbols now support "Parent.child" notation (e.g.,
   "Session.request") by matching against qualified names (::Parent::child).
   Previously only checked node.name === symbol, which never matched qualified
   queries since node names are unqualified.

Also adds Alamofire Swift benchmark data to README (91% fewer tool calls,
78% faster with CodeGraph).

Co-Authored-By: Claude Opus 4.6 (1M context) &lt;noreply@anthropic.com&gt;
diff --git a/README.md b/README.md
@@ -44,6 +44,7 @@ We tested the same exploration queries across 4 real-world codebases in differen
 | **Excalidraw** | TypeScript | "How does collaborative editing and real-time sync work?" | 3 calls, 29s | 47 calls, 1m 45s | **94% fewer** | **72% faster** |
 | **Claude Code** | Python + Rust | "How does tool execution work end to end?" | 3 calls, 39s | 40 calls, 1m 8s | **93% fewer** | **43% faster** |
 | **Claude Code** | Java | "How does tool execution work end to end?" | 1 call, 19s | 26 calls, 1m 22s | **96% fewer** | **77% faster** |
+| **Alamofire** | Swift | "Trace how a request flows from Session.request() through to the URLSession layer" | 3 calls, 22s | 32 calls, 1m 39s | **91% fewer** | **78% faster** |
 
 <details>
 <summary><strong>Full benchmark details</strong></summary>
@@ -57,6 +58,7 @@ All tests used Claude Opus 4.6 (1M context) with Claude Code v2.1.91. Each test
 | Excalidraw (TypeScript) | 626 | 9,859 | 3 | 57.1k | 29s | 0 |
 | Claude Code (Python+Rust) | 115 | 3,080 | 3 | 67.1k | 39s | 0 |
 | Claude Code (Java) | — | — | 1 | 40.8k | 19s | 0 |
+| Alamofire (Swift) | 102 | 2,624 | 3 | 57.3k | 22s | 0 |
 
 **Without CodeGraph — the agent uses grep, find, ls, and Read extensively:**
 | Codebase | Tool Uses | Tokens | Time | File Reads |
@@ -65,12 +67,14 @@ All tests used Claude Opus 4.6 (1M context) with Claude Code v2.1.91. Each test
 | Excalidraw (TypeScript) | 47 | 77.9k | 1m 45s | ~20 |
 | Claude Code (Python+Rust) | 40 | 69.3k | 1m 8s | ~15 |
 | Claude Code (Java) | 26 | 73.3k | 1m 22s | ~15 |
+| Alamofire (Swift) | 32 | 52.4k | 1m 39s | ~10 |
 
 **Key observations:**
 - With CodeGraph, the agent **never fell back to reading files** — it trusted the codegraph_explore results completely
 - Without CodeGraph, agents spent most of their time on discovery (find, ls, grep) before they could even start reading relevant code
 - The Java codebase needed only **1 codegraph_explore call** to answer the entire question
 - Cross-language queries (Python+Rust) worked seamlessly — CodeGraph's graph traversal found connections across language boundaries
+- The Swift benchmark (Alamofire) traced a **9-step call chain** from `Session.request()` to `URLSession.dataTask()` — CodeGraph's graph traversal at depth 3 captured the full chain in one explore call
 
 </details>
 
diff --git a/src/mcp/tools.ts b/src/mcp/tools.ts
@@ -622,7 +622,7 @@ export class ToolHandler {
     // Step 1: Find relevant context with generous parameters
     const subgraph = await cg.findRelevantContext(query, {
       searchLimit: 8,
-      traversalDepth: 2,
+      traversalDepth: 3,
       maxNodes: 80,
       minScore: 0.2,
     });
@@ -1113,15 +1113,38 @@ export class ToolHandler {
    * Find a symbol by name, handling disambiguation when multiple matches exist.
    * Returns the best match and a note about alternatives if any.
    */
+  /**
+   * Check if a node matches a symbol query, supporting both simple names and
+   * qualified "Parent.child" notation (e.g., "Session.request" matches a method
+   * named "request" inside a class named "Session").
+   */
+  private matchesSymbol(node: Node, symbol: string): boolean {
+    // Simple name match
+    if (node.name === symbol) return true;
+    // File basename match (e.g., "product-card" matches "product-card.liquid")
+    if (node.kind === 'file' && node.name.replace(/\.[^.]+$/, '') === symbol) return true;
+
+    // Qualified name match: "Parent.child" → look for "::Parent::child" in qualified_name
+    if (symbol.includes('.')) {
+      const parts = symbol.split('.');
+      const qualifiedSuffix = parts.join('::');
+      if (node.qualifiedName.includes(qualifiedSuffix)) return true;
+    }
+
+    return false;
+  }
+
   private findSymbol(cg: CodeGraph, symbol: string): { node: Node; note: string } | null {
-    const results = cg.searchNodes(symbol, { limit: 10 });
+    // Use higher limit for qualified lookups (e.g., "Session.request") since the
+    // target may rank lower in FTS when there are many partial matches
+    const limit = symbol.includes('.') ? 50 : 10;
+    const results = cg.searchNodes(symbol, { limit });
 
     if (results.length === 0 || !results[0]) {
       return null;
     }
 
-    // If only one result, or first is an exact name match, use it directly
-    const exactMatches = results.filter(r => r.node.name === symbol);
+    const exactMatches = results.filter(r => this.matchesSymbol(r.node, symbol));
 
     if (exactMatches.length === 1) {
       return { node: exactMatches[0]!.node, note: '' };
@@ -1152,12 +1175,7 @@ export class ToolHandler {
       return { nodes: [], note: '' };
     }
 
-    // Match by exact name, OR by file basename without extension
-    // (e.g., "product-card" matches file node named "product-card.liquid")
-    const exactMatches = results.filter(r =>
-      r.node.name === symbol ||
-      (r.node.kind === 'file' && r.node.name.replace(/\.[^.]+$/, '') === symbol)
-    );
+    const exactMatches = results.filter(r => this.matchesSymbol(r.node, symbol));
 
     if (exactMatches.length <= 1) {
       const node = exactMatches[0]?.node ?? results[0]!.node;
