@comfanion/usethis_search 3.0.0-dev.18 → 3.0.0-dev.19

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@comfanion/usethis_search",
-  "version": "3.0.0-dev.18",
+  "version": "3.0.0-dev.19",
   "description": "OpenCode plugin: semantic search with graph-based context (v3: graph relations, 1-hop context, LSP + regex analyzers)",
   "type": "module",
   "main": "./index.ts",

package/vectorizer/index.ts

@@ -90,13 +90,15 @@ const DEFAULT_GRAPH_CONFIG = {
   enabled: true,
   max_related: 4,
   min_relevance: 0.5,
+  semantic_edges: false,          // O(n²) post-pass — disabled by default (slow on large repos)
+  semantic_edges_max_chunks: 500, // Skip semantic edges if chunk count exceeds this
   lsp: {
     enabled: true,
     timeout_ms: 5000,
   },
   read_intercept: true,
 };
-let GRAPH_CONFIG = { ...DEFAULT_GRAPH_CONFIG, lsp: { ...DEFAULT_GRAPH_CONFIG.lsp } };
+let GRAPH_CONFIG = { ...DEFAULT_GRAPH_CONFIG, lsp: { ...DEFAULT_GRAPH_CONFIG.lsp } } as typeof DEFAULT_GRAPH_CONFIG;
 
 function defaultVectorizerYaml() {
   return (
@@ -139,6 +141,8 @@ function defaultVectorizerYaml() {
     `    enabled: true\n` +
     `    max_related: 4\n` +
     `    min_relevance: 0.5\n` +
+    `    semantic_edges: false        # O(n²) post-pass — enable only for small repos\n` +
+    `    semantic_edges_max_chunks: 500  # Skip if chunk count exceeds this\n` +
     `    lsp:\n` +
     `      enabled: true\n` +
     `      timeout_ms: 5000\n` +
@@ -312,6 +316,8 @@ async function loadConfig(projectRoot) {
       GRAPH_CONFIG.enabled = parseBool(gs, "enabled", DEFAULT_GRAPH_CONFIG.enabled);
       GRAPH_CONFIG.max_related = parseNumber(gs, "max_related", DEFAULT_GRAPH_CONFIG.max_related);
       GRAPH_CONFIG.min_relevance = parseNumber(gs, "min_relevance", DEFAULT_GRAPH_CONFIG.min_relevance);
+      GRAPH_CONFIG.semantic_edges = parseBool(gs, "semantic_edges", DEFAULT_GRAPH_CONFIG.semantic_edges);
+      GRAPH_CONFIG.semantic_edges_max_chunks = parseNumber(gs, "semantic_edges_max_chunks", DEFAULT_GRAPH_CONFIG.semantic_edges_max_chunks);
       GRAPH_CONFIG.read_intercept = parseBool(gs, "read_intercept", DEFAULT_GRAPH_CONFIG.read_intercept);
 
       // Nested lsp: section
@@ -1093,9 +1099,9 @@ class CodebaseIndexer {
     }
 
     // FR-005: Build semantic similarity edges as post-pass
-    // Only if we actually indexed new files and have a graph builder
+    // Disabled by default (O(n²) — slow on large repos). Enable via graph.semantic_edges: true
     let semanticEdges = 0;
-    if (indexed > 0 && this.graphBuilder && this.graphDB) {
+    if (indexed > 0 && this.graphBuilder && this.graphDB && GRAPH_CONFIG.semantic_edges) {
       try {
         const tableName = "chunks";
         const tables = await this.db.tableNames();
@@ -1105,8 +1111,15 @@ class CodebaseIndexer {
           const chunkData = allRows
             .filter(r => r.chunk_id && r.vector)
             .map(r => ({ chunk_id: r.chunk_id, vector: Array.from(r.vector), file: r.file }));
-          semanticEdges = await this.graphBuilder.buildSemanticEdges(chunkData, 0.8, 3);
-          if (DEBUG) console.log(`[vectorizer] Built ${semanticEdges} semantic similarity edges`);
+
+          // Skip if too many chunks — O(n²) becomes prohibitive
+          const maxChunks = GRAPH_CONFIG.semantic_edges_max_chunks ?? 500;
+          if (chunkData.length > maxChunks) {
+            if (DEBUG) console.log(`[vectorizer] Skipping semantic edges: ${chunkData.length} chunks > max ${maxChunks}`);
+          } else {
+            semanticEdges = await this.graphBuilder.buildSemanticEdges(chunkData, 0.8, 3);
+            if (DEBUG) console.log(`[vectorizer] Built ${semanticEdges} semantic similarity edges`);
+          }
         }
       } catch (e) {
         if (DEBUG) console.log(`[vectorizer] Semantic edge building failed:`, e.message);

package/vectorizer.yaml

@@ -44,6 +44,8 @@ vectorizer:
     enabled: true
     max_related: 4           # How many related chunks to attach
     min_relevance: 0.5       # Minimum score threshold for related context
+    semantic_edges: false    # O(n²) post-pass — enable only for small repos (<500 chunks)
+    semantic_edges_max_chunks: 500  # Skip semantic edges if chunk count exceeds this
 
     # LSP for code analysis
     lsp: