@ez-corp/ez-search 1.0.3 → 1.0.5

package/dist/cli/commands/index-cmd.js

@@ -235,6 +235,8 @@ export async function runIndex(targetPath, options) {
         // 3. Handle --clear
         // rmSync removes .ez-search/ entirely (including manifest.json inside it)
         if (options.clear) {
+            col768.close();
+            col512.close();
             rmSync(storagePath, { recursive: true, force: true });
             const reopened = openProjectCollections(absPath);
             col768 = reopened.col768;
@@ -390,11 +392,13 @@ export async function runIndex(targetPath, options) {
             const format = options.format === 'text' ? 'text' : 'json';
             emitError({ code: 'EMPTY_DIR', message: 'No supported files found in directory', suggestion: 'Ensure the directory contains supported file types (.ts, .js, .py, .go, .rs, .c, .cpp, .md, .txt, .jpg, .png, .webp)' }, format);
         }
-        // 6. Optimize THEN save manifest
+        // 6. Optimize, close collections, THEN save manifest
         col768.optimize();
+        col768.close();
         if (imageFilesProcessed) {
             col512.optimize();
         }
+        col512.close();
         saveManifest(absPath, manifest);
         // 7. Output results
         const durationMs = Date.now() - startTime;

package/dist/cli/commands/query-cmd.js

@@ -53,176 +53,181 @@ export async function runQuery(text, options) {
             staleFileCount = await calcStaleness(projectDir, manifest, true);
         }
         const isStale = staleFileCount > 0;
-        // 3. Open vector collections (after auto-index so DB files exist)
-        const { openProjectCollections } = await import('../../services/vector-db.js');
-        const { col768 } = openProjectCollections(projectDir);
-        let typesToQuery;
-        if (options.type) {
-            typesToQuery = [options.type];
-        }
-        else {
-            // Pre-detect indexed types from manifest: only load models for types that have data.
-            // This avoids loading Jina when only text is indexed (or Nomic when only code is indexed).
-            const { EXTENSION_MAP } = await import('../../types.js');
-            const indexedTypes = new Set();
-            for (const filePath of Object.keys(manifest.files)) {
-                const ext = '.' + filePath.split('.').pop()?.toLowerCase();
-                const fileType = EXTENSION_MAP[ext];
-                if (fileType)
-                    indexedTypes.add(fileType);
+        // 3. Open only the text/code collection (query never needs col-512 for images)
+        const { openCollection } = await import('../../services/vector-db.js');
+        const col768 = openCollection(projectDir, 'col-768');
+        try {
+            let typesToQuery;
+            if (options.type) {
+                typesToQuery = [options.type];
+            }
+            else {
+                // Pre-detect indexed types from manifest: only load models for types that have data.
+                // This avoids loading Jina when only text is indexed (or Nomic when only code is indexed).
+                const { EXTENSION_MAP } = await import('../../types.js');
+                const indexedTypes = new Set();
+                for (const filePath of Object.keys(manifest.files)) {
+                    const ext = '.' + filePath.split('.').pop()?.toLowerCase();
+                    const fileType = EXTENSION_MAP[ext];
+                    if (fileType)
+                        indexedTypes.add(fileType);
+                }
+                typesToQuery = [];
+                if (indexedTypes.has('code'))
+                    typesToQuery.push('code');
+                if (indexedTypes.has('text'))
+                    typesToQuery.push('text');
+                // image queries from text not supported — skip even if images are indexed
+            }
+            // Early exit when manifest exists but has no queryable types (e.g., after --clear without re-indexing)
+            if (typesToQuery.length === 0) {
+                const { emitError } = await import('../errors.js');
+                emitError({ code: 'NO_INDEX', message: 'No indexed content found', suggestion: 'Run `ez-search index .` first' }, options.format === 'text' ? 'text' : 'json');
             }
-            typesToQuery = [];
-            if (indexedTypes.has('code'))
-                typesToQuery.push('code');
-            if (indexedTypes.has('text'))
-                typesToQuery.push('text');
-            // image queries from text not supported — skip even if images are indexed
-        }
-        // Early exit when manifest exists but has no queryable types (e.g., after --clear without re-indexing)
-        if (typesToQuery.length === 0) {
-            const { emitError } = await import('../errors.js');
-            emitError({ code: 'NO_INDEX', message: 'No indexed content found', suggestion: 'Run `ez-search index .` first' }, options.format === 'text' ? 'text' : 'json');
-        }
-        // Handle unsupported image query
-        if (options.type === 'image') {
-            const { emitError } = await import('../errors.js');
-            emitError({
-                code: 'UNSUPPORTED_TYPE',
-                message: 'Image search requires image query input (not yet supported)',
-                suggestion: 'Omit --type image to search code and text',
-            }, options.format === 'text' ? 'text' : 'json');
-        }
-        // ── Helpers ──────────────────────────────────────────────────────────────
-        const { normalizeResults, filterAndCollapse } = await import('../../services/query-utils.js');
-        const hasPostFilters = options.dir !== undefined || threshold !== undefined;
-        // Over-fetch for mixed col-768 + optional post-filters
-        const fetchCount = topK * 5 * (hasPostFilters ? 3 : 1);
-        // ── Execute per-type queries sequentially (memory conservation) ──────────
-        const { createEmbeddingPipeline } = await import('../../services/model-router.js');
-        let codeResults = [];
-        let textResults = [];
-        if (typesToQuery.includes('code')) {
-            // Code: Jina embedding, filter for jina modelId
-            let pipe = null;
-            try {
-                pipe = await createEmbeddingPipeline('code');
-                const [queryEmbedding] = await pipe.embed([text]);
-                let rawResults;
+            // Handle unsupported image query
+            if (options.type === 'image') {
+                const { emitError } = await import('../errors.js');
+                emitError({
+                    code: 'UNSUPPORTED_TYPE',
+                    message: 'Image search requires image query input (not yet supported)',
+                    suggestion: 'Omit --type image to search code and text',
+                }, options.format === 'text' ? 'text' : 'json');
+            }
+            // ── Helpers ──────────────────────────────────────────────────────────────
+            const { normalizeResults, filterAndCollapse } = await import('../../services/query-utils.js');
+            const hasPostFilters = options.dir !== undefined || threshold !== undefined;
+            // Over-fetch for mixed col-768 + optional post-filters
+            const fetchCount = topK * 5 * (hasPostFilters ? 3 : 1);
+            // ── Execute per-type queries sequentially (memory conservation) ──────────
+            const { createEmbeddingPipeline } = await import('../../services/model-router.js');
+            let codeResults = [];
+            let textResults = [];
+            if (typesToQuery.includes('code')) {
+                // Code: Jina embedding, filter for jina modelId
+                let pipe = null;
                 try {
-                    rawResults = col768.query(queryEmbedding, fetchCount);
+                    pipe = await createEmbeddingPipeline('code');
+                    const [queryEmbedding] = await pipe.embed([text]);
+                    let rawResults;
+                    try {
+                        rawResults = col768.query(queryEmbedding, fetchCount);
+                    }
+                    catch {
+                        rawResults = [];
+                    }
+                    const normalized = normalizeResults(rawResults);
+                    codeResults = filterAndCollapse(normalized, (id) => id.includes('jina') || id.startsWith('jinaai/'), { threshold, dir: options.dir, topK });
                 }
-                catch {
-                    rawResults = [];
+                catch (err) {
+                    process.stderr.write(`[query] code pipeline error: ${err instanceof Error ? err.message : String(err)}\n`);
+                }
+                finally {
+                    if (pipe)
+                        await pipe.dispose();
                 }
-                const normalized = normalizeResults(rawResults);
-                codeResults = filterAndCollapse(normalized, (id) => id.includes('jina') || id.startsWith('jinaai/'), { threshold, dir: options.dir, topK });
-            }
-            catch (err) {
-                process.stderr.write(`[query] code pipeline error: ${err instanceof Error ? err.message : String(err)}\n`);
-            }
-            finally {
-                if (pipe)
-                    await pipe.dispose();
             }
-        }
-        if (typesToQuery.includes('text')) {
-            // Text: Nomic embedding with "search_query: " prefix, filter for nomic modelId
-            let pipe = null;
-            try {
-                pipe = await createEmbeddingPipeline('text');
-                const prefixedQuery = `search_query: ${text}`;
-                const [queryEmbedding] = await pipe.embed([prefixedQuery]);
-                let rawResults;
+            if (typesToQuery.includes('text')) {
+                // Text: Nomic embedding with "search_query: " prefix, filter for nomic modelId
+                let pipe = null;
                 try {
-                    rawResults = col768.query(queryEmbedding, fetchCount);
+                    pipe = await createEmbeddingPipeline('text');
+                    const prefixedQuery = `search_query: ${text}`;
+                    const [queryEmbedding] = await pipe.embed([prefixedQuery]);
+                    let rawResults;
+                    try {
+                        rawResults = col768.query(queryEmbedding, fetchCount);
+                    }
+                    catch {
+                        rawResults = [];
+                    }
+                    const normalized = normalizeResults(rawResults);
+                    textResults = filterAndCollapse(normalized, (id) => id.includes('nomic'), { threshold, dir: options.dir, topK });
                 }
-                catch {
-                    rawResults = [];
+                catch (err) {
+                    process.stderr.write(`[query] text pipeline error: ${err instanceof Error ? err.message : String(err)}\n`);
+                }
+                finally {
+                    if (pipe)
+                        await pipe.dispose();
                 }
-                const normalized = normalizeResults(rawResults);
-                textResults = filterAndCollapse(normalized, (id) => id.includes('nomic'), { threshold, dir: options.dir, topK });
-            }
-            catch (err) {
-                process.stderr.write(`[query] text pipeline error: ${err instanceof Error ? err.message : String(err)}\n`);
-            }
-            finally {
-                if (pipe)
-                    await pipe.dispose();
-            }
-        }
-        // ── Output ────────────────────────────────────────────────────────────────
-        const hasCodeResults = codeResults.length > 0;
-        const hasTextResults = textResults.length > 0;
-        const hasResults = hasCodeResults || hasTextResults;
-        if (options.format === 'text') {
-            if (autoIndexResult) {
-                console.log(`Auto-indexed ${autoIndexResult.filesIndexed} files in ${(autoIndexResult.durationMs / 1000).toFixed(1)}s\n`);
-            }
-            if (isStale) {
-                console.log(`Warning: ${staleFileCount} file(s) changed since last index. Run \`ez-search index .\` to update.\n`);
-            }
-            if (!hasResults) {
-                console.log('No results found.');
-                return;
             }
-            if (hasCodeResults) {
-                console.log('## Code\n');
-                for (const r of codeResults) {
-                    console.log(`File: ${r.filePath} | Lines: ${r.lineStart}-${r.lineEnd} | Relevance: ${r.score}`);
-                    for (const line of r.chunkText.split('\n')) {
-                        console.log(`    ${line}`);
+            // ── Output ────────────────────────────────────────────────────────────────
+            const hasCodeResults = codeResults.length > 0;
+            const hasTextResults = textResults.length > 0;
+            const hasResults = hasCodeResults || hasTextResults;
+            if (options.format === 'text') {
+                if (autoIndexResult) {
+                    console.log(`Auto-indexed ${autoIndexResult.filesIndexed} files in ${(autoIndexResult.durationMs / 1000).toFixed(1)}s\n`);
+                }
+                if (isStale) {
+                    console.log(`Warning: ${staleFileCount} file(s) changed since last index. Run \`ez-search index .\` to update.\n`);
+                }
+                if (!hasResults) {
+                    console.log('No results found.');
+                    return;
+                }
+                if (hasCodeResults) {
+                    console.log('## Code\n');
+                    for (const r of codeResults) {
+                        console.log(`File: ${r.filePath} | Lines: ${r.lineStart}-${r.lineEnd} | Relevance: ${r.score}`);
+                        for (const line of r.chunkText.split('\n')) {
+                            console.log(`    ${line}`);
+                        }
+                        console.log();
                     }
-                    console.log();
                 }
-            }
-            if (hasTextResults) {
-                console.log('## Text\n');
-                for (const r of textResults) {
-                    console.log(`File: ${r.filePath} | Relevance: ${r.score}`);
-                    for (const line of r.chunkText.split('\n')) {
-                        console.log(`    ${line}`);
+                if (hasTextResults) {
+                    console.log('## Text\n');
+                    for (const r of textResults) {
+                        console.log(`File: ${r.filePath} | Relevance: ${r.score}`);
+                        for (const line of r.chunkText.split('\n')) {
+                            console.log(`    ${line}`);
+                        }
+                        console.log();
                     }
-                    console.log();
                 }
             }
-        }
-        else {
-            // JSON grouped envelope
-            const output = {
-                query: text,
-                totalIndexed,
-                searchScope: options.dir ?? '.',
-            };
-            if (autoIndexResult) {
-                output['indexing'] = {
-                    status: autoIndexResult.status,
-                    filesIndexed: autoIndexResult.filesIndexed,
-                    durationMs: autoIndexResult.durationMs,
+            else {
+                // JSON grouped envelope
+                const output = {
+                    query: text,
+                    totalIndexed,
+                    searchScope: options.dir ?? '.',
                 };
+                if (autoIndexResult) {
+                    output['indexing'] = {
+                        status: autoIndexResult.status,
+                        filesIndexed: autoIndexResult.filesIndexed,
+                        durationMs: autoIndexResult.durationMs,
+                    };
+                }
+                if (isStale) {
+                    output['stale'] = true;
+                    output['staleFileCount'] = staleFileCount;
+                }
+                if (hasCodeResults) {
+                    output['code'] = codeResults.map((r) => ({
+                        file: r.filePath,
+                        lines: { start: r.lineStart, end: r.lineEnd },
+                        score: r.score,
+                        text: r.chunkText,
+                    }));
+                }
+                if (hasTextResults) {
+                    output['text'] = textResults.map((r) => ({
+                        file: r.filePath,
+                        score: r.score,
+                        text: r.chunkText,
+                    }));
+                }
+                if (!hasResults) {
+                    output['message'] = 'No results found';
+                }
+                console.log(JSON.stringify(output));
             }
-            if (isStale) {
-                output['stale'] = true;
-                output['staleFileCount'] = staleFileCount;
-            }
-            if (hasCodeResults) {
-                output['code'] = codeResults.map((r) => ({
-                    file: r.filePath,
-                    lines: { start: r.lineStart, end: r.lineEnd },
-                    score: r.score,
-                    text: r.chunkText,
-                }));
-            }
-            if (hasTextResults) {
-                output['text'] = textResults.map((r) => ({
-                    file: r.filePath,
-                    score: r.score,
-                    text: r.chunkText,
-                }));
-            }
-            if (!hasResults) {
-                output['message'] = 'No results found';
-            }
-            console.log(JSON.stringify(output));
+        }
+        finally {
+            col768.close();
         }
     }
     catch (err) {

package/dist/services/vector-db.js

@@ -101,7 +101,11 @@ function createCollection(storageDir, name, dim) {
                     chunkText: String(metadata['chunkText'] ?? ''),
                 },
             };
-            const status = handle.insertSync(doc);
+            let status = handle.insertSync(doc);
+            if (!status.ok && status.code === 'ZVEC_ALREADY_EXISTS') {
+                handle.deleteSync(id);
+                status = handle.insertSync(doc);
+            }
             if (!status.ok) {
                 throw new Error(`Zvec insert failed for id="${id}": code=${status.code} ${status.message}`);
             }
@@ -136,9 +140,7 @@ function createCollection(storageDir, name, dim) {
             handle.optimizeSync();
         },
         close() {
-            // Zvec collections are closed/cleaned up when the handle goes out of scope.
-            // Call destroySync only if you want to delete the data from disk.
-            // For normal close, we do nothing — the GC handles cleanup.
+            handle.closeSync();
         },
     };
 }
@@ -159,3 +161,14 @@ export function openProjectCollections(projectDir) {
     const col512 = createCollection(storageDir, 'col-512', 512);
     return { col768, col512, storagePath: storageDir };
 }
+/**
+ * Open a single vector collection by name.
+ * Use this when you only need one collection (e.g. query only needs col-768).
+ */
+export function openCollection(projectDir, name) {
+    const storageDir = resolveProjectStoragePath(projectDir);
+    mkdirSync(storageDir, { recursive: true });
+    ensureSchemaVersion(storageDir);
+    const dim = name === 'col-768' ? 768 : 512;
+    return createCollection(storageDir, name, dim);
+}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ez-corp/ez-search",
-  "version": "1.0.3",
+  "version": "1.0.5",
   "description": "Semantic codebase search with zero cloud dependencies",
   "type": "module",
   "main": "dist/cli/index.js",