eminiar 0.10.0 → 0.10.1

package/dist/hook.js

@@ -26434,7 +26434,7 @@ function chunkMarkdown(text, baseTitle, opts = DEFAULT_CHUNKER) {
     if (/^\s*(```|~~~)/.test(line2)) inFence = !inFence;
     const m = !inFence && /^(#{1,6})\s+(.+?)\s*#*\s*$/.exec(line2);
     if (m) {
-      if (current.lines.length) sections.push(current);
+      sections.push(current);
       const level = m[1].length;
       while (stack.length && stack[stack.length - 1].level >= level) stack.pop();
       stack.push({ level, text: m[2] });
@@ -26443,14 +26443,23 @@ function chunkMarkdown(text, baseTitle, opts = DEFAULT_CHUNKER) {
     }
     current.lines.push(line2);
   });
-  if (current.lines.length) sections.push(current);
+  sections.push(current);
   const out = [];
-  for (const s of sections) {
+  sections.forEach((s, k) => {
     const title = s.path.length ? `${baseTitle} > ${s.path.join(" > ")}` : baseTitle;
     const body = s.lines.join("\n");
+    if (!body.trim()) {
+      const next = sections[k + 1];
+      const isParent = next !== void 0 && next.path.length > s.path.length && s.path.every((p, j) => next.path[j] === p);
+      if (s.path.length && !isParent) {
+        const heading = s.path[s.path.length - 1];
+        out.push({ title, body: heading, startLine: s.startLine, endLine: s.startLine });
+      }
+      return;
+    }
     if (body.length <= opts.maxChars) {
       out.push({ title, body, startLine: s.startLine, endLine: s.startLine + s.lines.length - 1 });
-      continue;
+      return;
     }
     let fence = false;
     const fenceState = s.lines.map((l) => {
@@ -26466,7 +26475,7 @@ function chunkMarkdown(text, baseTitle, opts = DEFAULT_CHUNKER) {
         s.startLine - 1
       )
     );
-  }
+  });
   return out;
 }
 
@@ -26549,7 +26558,20 @@ function chunk(content, baseTitle, kind, uri, opts = DEFAULT_CHUNKER) {
 // src/index/indexer.ts
 function indexContent(db, input2) {
   const kind = input2.kind ?? (input2.contentType === "code" ? "code" : input2.contentType === "json" ? "json" : void 0);
-  const { chunks, kind: k } = input2.chunks ? { chunks: finalize2(input2.chunks, input2.contentType ?? "code"), kind: kind ?? "code" } : chunk(input2.content, input2.title, kind, input2.uri);
+  const cut = input2.chunks ? { chunks: finalize2(input2.chunks, input2.contentType ?? "code"), kind: kind ?? "code" } : chunk(input2.content, input2.title, kind, input2.uri);
+  const k = cut.kind;
+  const fallback = cut.chunks.length === 0 && input2.content.trim().length > 0;
+  const chunks = fallback ? finalize2(
+    [
+      {
+        title: input2.title,
+        body: input2.content,
+        startLine: 1,
+        endLine: input2.content.split("\n").length
+      }
+    ],
+    kindToContentType(k)
+  ) : cut.chunks;
   const contentHash = hashText(input2.content);
   const existing = input2.uri ? db.findDocumentByUri(input2.uri) : void 0;
   if (existing && existing.content_hash === contentHash) {
@@ -26559,7 +26581,8 @@ function indexContent(db, input2) {
         created: false,
         chunks: chunks.length,
         chars: input2.content.length,
-        kind: k
+        kind: k,
+        fallback
       };
     }
     db.deleteDocument(existing.id);
@@ -26576,7 +26599,14 @@ function indexContent(db, input2) {
     },
     chunks
   );
-  return { documentId: id, created, chunks: chunks.length, chars: input2.content.length, kind: k };
+  return {
+    documentId: id,
+    created,
+    chunks: chunks.length,
+    chars: input2.content.length,
+    kind: k,
+    fallback
+  };
 }
 var MAX_FILE_BYTES = 2 * 1024 * 1024;
 
@@ -26597,6 +26627,23 @@ function fileOf(id) {
   return void 0;
 }
 
+// src/graph/metrics.ts
+init_esm_shims();
+
+// src/graph/clusters.ts
+init_esm_shims();
+
+// src/graph/metrics.ts
+var DEPENDENCY_KINDS = [
+  "calls",
+  "references",
+  "extends",
+  "implements",
+  "imports",
+  "reexports",
+  "handles"
+];
+
 // src/history/model.ts
 init_esm_shims();
 
@@ -27298,6 +27345,9 @@ function memoriesForFile(db, rel, opts = {}) {
 }
 var RISKY = 0.3;
 var FAN_IN = 5;
+function dependentFiles(db, paths2) {
+  return db.graphDependentFileCounts(paths2, DEPENDENCY_KINDS);
+}
 function riskLine(path2, f) {
   if (f.risk < RISKY && !f.bugOrigins) return void 0;
   const parts = [`${f.fixes}/${f.commits} commits were fixes`];
@@ -27325,10 +27375,10 @@ function fileContext(db, rel, opts = {}) {
   }
   const fanKey = `fanin:${rel}`;
   if (opts.edit && !seen.has(fanKey)) {
-    const fanIn = db.metricsGet([fileId(rel)])[0]?.fan_in ?? 0;
-    if (fanIn >= FAN_IN) {
+    const dependents = dependentFiles(db, [rel]).get(rel) ?? 0;
+    if (dependents >= FAN_IN) {
       lines.push(
-        `eminiar graph: ${rel} is imported by ${fanIn} files; em_graph_impact {node:"${fileId(rel)}"} lists them before you change an exported signature.`
+        `eminiar graph: ${rel} is imported by ${dependents} files; em_graph_impact {node:"${fileId(rel)}"} lists them before you change an exported signature.`
       );
       keys.push(fanKey);
     }
@@ -27340,7 +27390,7 @@ function commitContext(db, files) {
   if (!files.length) return "";
   const table = readFileRisk(db);
   const risky = files.map((p) => table[p] ? riskLine(p, table[p]) : void 0).filter((l) => !!l).slice(0, 4);
-  const wide = db.metricsGet(files.map(fileId)).filter((m) => m.fan_in >= FAN_IN).sort((a, b) => b.fan_in - a.fan_in).slice(0, 4).map((m) => `${m.path} (${m.fan_in} importers)`);
+  const wide = [...dependentFiles(db, files)].filter(([, n]) => n >= FAN_IN).sort((a, b) => b[1] - a[1]).slice(0, 4).map(([p, n]) => `${p} (${n} importers)`);
   if (!risky.length && !wide.length) return "";
   const lines = ["eminiar history: this commit includes files that deserve a second look:"];
   for (const r of risky) lines.push(`- ${r}`);
@@ -27624,6 +27674,10 @@ function topTermsOf(bodies, limit) {
 
 // src/storage/sqlite/schema.ts
 init_esm_shims();
+function hasColumn(db, table, column) {
+  const cols = db.pragma(`table_info(${table})`);
+  return cols.some((c) => c.name === column);
+}
 var migrations = [
   {
     version: 1,
@@ -27958,7 +28012,6 @@ CREATE TABLE IF NOT EXISTS transcript_files (
   offset INTEGER NOT NULL DEFAULT 0,
   size INTEGER NOT NULL DEFAULT 0,
   mtime INTEGER NOT NULL DEFAULT 0,
-  head_hash TEXT,
   ingested_at INTEGER NOT NULL
 );
 
@@ -27986,6 +28039,16 @@ CREATE TABLE IF NOT EXISTS statusline_snapshots (
 CREATE INDEX IF NOT EXISTS statusline_snapshots_session_ts ON statusline_snapshots(session_id, ts);
 CREATE INDEX IF NOT EXISTS statusline_snapshots_ts ON statusline_snapshots(ts);
 `
+  },
+  {
+    // transcript_files.head_hash (fingerprint of the file head, so a rewrite with the same
+    // size is noticed) was first added to v5 after v5 had already run on real databases, so
+    // version-5 databases exist both with and without the column. Add it only when missing.
+    version: 6,
+    up(db) {
+      if (!hasColumn(db, "transcript_files", "head_hash"))
+        db.exec("ALTER TABLE transcript_files ADD COLUMN head_hash TEXT");
+    }
   }
 ];
 function vecTableSql(dims) {
@@ -28097,7 +28160,8 @@ var SqliteAdapter = class _SqliteAdapter {
     if (pending.length) {
       this.transaction(() => {
         for (const m of pending) {
-          this.db.exec(m.sql);
+          if (m.sql) this.db.exec(m.sql);
+          m.up?.(this.db);
           this.metaSet("schema_version", String(m.version));
         }
       });
@@ -28316,18 +28380,42 @@ var SqliteAdapter = class _SqliteAdapter {
   }
   vecSearch(embedding, k, filters) {
     if (!this.vectors) return [];
-    const raw = this.db.prepare(
-      "SELECT chunk_id id, distance FROM chunk_vec WHERE embedding MATCH ? AND k = ? ORDER BY distance"
-    ).all(embedding, BigInt(filters ? k * 4 : k));
-    if (!filters || !raw.length) return raw.map((r) => ({ chunkId: r.id, distance: r.distance }));
     const { where, params } = filterSql(filters);
-    const allowed = new Set(
-      this.db.prepare(
-        `SELECT c.id FROM chunks c JOIN documents d ON d.id = c.document_id
-             WHERE c.id IN (${raw.map(() => "?").join(",")})${where}`
-      ).all(...raw.map((r) => r.id), ...params).map((r) => r.id)
-    );
-    return raw.filter((r) => allowed.has(r.id)).slice(0, k).map((r) => ({ chunkId: r.id, distance: r.distance }));
+    if (!where) return this.knn(embedding, k);
+    try {
+      return this.knn(
+        embedding,
+        k,
+        `AND chunk_id IN (SELECT c.id FROM chunks c JOIN documents d ON d.id = c.document_id WHERE 1=1${where})`,
+        params
+      );
+    } catch {
+      return this.vecSearchWidening(embedding, k, where, params);
+    }
+  }
+  knn(embedding, k, extra = "", params = []) {
+    return this.db.prepare(
+      `SELECT chunk_id id, distance FROM chunk_vec WHERE embedding MATCH ? AND k = ? ${extra} ORDER BY distance`
+    ).all(embedding, BigInt(k), ...params).map((r) => ({ chunkId: r.id, distance: r.distance }));
+  }
+  /** Fallback filtered KNN: fetch k*4, k*16, ... nearest until k pass the filter or all were seen. */
+  vecSearchWidening(embedding, k, where, params) {
+    const total = this.vecCount();
+    for (let fetch = k * 4; ; fetch *= 4) {
+      const raw = this.knn(embedding, Math.min(fetch, total));
+      if (!raw.length) return [];
+      const allowed = /* @__PURE__ */ new Set();
+      for (let i = 0; i < raw.length; i += 400) {
+        const ids = raw.slice(i, i + 400).map((r) => r.chunkId);
+        for (const r of this.db.prepare(
+          `SELECT c.id FROM chunks c JOIN documents d ON d.id = c.document_id
+             WHERE c.id IN (${ids.map(() => "?").join(",")})${where}`
+        ).all(...ids, ...params))
+          allowed.add(r.id);
+      }
+      const hits = raw.filter((r) => allowed.has(r.chunkId)).slice(0, k);
+      if (hits.length >= k || fetch >= total) return hits;
+    }
   }
   unembeddedChunkIds(limit) {
     return this.db.prepare("SELECT id FROM chunks WHERE embedded = 0 ORDER BY id LIMIT ?").all(limit).map((r) => r.id);
@@ -28606,6 +28694,26 @@ var SqliteAdapter = class _SqliteAdapter {
     }
     return out;
   }
+  graphDependentFileCounts(paths2, kinds) {
+    const out = /* @__PURE__ */ new Map();
+    if (!paths2.length || !kinds.length) return out;
+    const kph = kinds.map(() => "?").join(",");
+    for (let i = 0; i < paths2.length; i += 400) {
+      const batch = paths2.slice(i, i + 400);
+      const pph = batch.map(() => "?").join(",");
+      for (const r of this.db.prepare(
+        `SELECT d.path p, COUNT(DISTINCT s.path) n
+             FROM graph_nodes d
+             JOIN graph_edges e ON e.dst = d.id AND e.kind IN (${kph})
+             JOIN graph_nodes s ON s.id = e.src
+            WHERE d.path IN (${pph}) AND s.path IS NOT NULL AND s.path <> d.path
+            GROUP BY d.path`
+      ).all(...kinds, ...batch)) {
+        out.set(r.p, r.n);
+      }
+    }
+    return out;
+  }
   edgesWhere(col, ids, kinds) {
     if (!ids.length) return [];
     const out = [];
@@ -30755,6 +30863,18 @@ function ingestSession(db, root, sessionId, transcriptPath, opts = {}) {
   return total;
 }
 
+// src/hooks/log.ts
+init_esm_shims();
+import { appendFileSync } from "fs";
+function hookLog(level, msg) {
+  try {
+    const p = ensureDirs();
+    appendFileSync(p.hookLog, `${(/* @__PURE__ */ new Date()).toISOString()} ${level} ${msg}
+`);
+  } catch {
+  }
+}
+
 // src/hooks/capture.ts
 var INDEXED = /* @__PURE__ */ new Set([
   "decision",
@@ -30797,6 +30917,7 @@ function open3(input2) {
   });
   return { db, root: ctx.root, sessionId };
 }
+var usageFailureLogged = false;
 function ingestUsage(db, input2, sessionId) {
   try {
     const ctx = getContext(input2.cwd);
@@ -30805,7 +30926,13 @@ function ingestUsage(db, input2, sessionId) {
     ingestSession(db, ctx.root, sessionId, input2.transcript_path, { overrides });
     if (input2.agent_transcript_path)
       ingestTranscript(db, input2.agent_transcript_path, { sessionId, overrides });
-  } catch {
+  } catch (err) {
+    if (usageFailureLogged) return;
+    usageFailureLogged = true;
+    hookLog(
+      "error",
+      `usage ingest failed (session ${sessionId}): ${err.message}; run \`eminiar usage ingest\` to see the error`
+    );
   }
 }
 var userPromptSubmit = async (input2) => {
@@ -31072,6 +31199,7 @@ import { extname as extname2, join as join12, relative as relative4, resolve as
 init_esm_shims();
 import { existsSync as existsSync4, mkdirSync as mkdirSync2, readFileSync as readFileSync3, writeFileSync } from "fs";
 import { dirname as dirname3, join as join5 } from "path";
+var DOCS_PREFIX = "docs:";
 var NAME_RE = /^[a-z0-9][a-z0-9._-]{0,63}$/;
 function isValidName(name) {
   return NAME_RE.test(name);
@@ -31451,18 +31579,6 @@ function effectiveRoutingMode(config2) {
   return contextModeInstalled() ? "annotate" : "enforce";
 }
 
-// src/hooks/log.ts
-init_esm_shims();
-import { appendFileSync } from "fs";
-function hookLog(level, msg) {
-  try {
-    const p = ensureDirs();
-    appendFileSync(p.hookLog, `${(/* @__PURE__ */ new Date()).toISOString()} ${level} ${msg}
-`);
-  } catch {
-  }
-}
-
 // src/hooks/routing.ts
 var SYMBOL_KINDS2 = [
   "function",
@@ -31573,6 +31689,7 @@ var BIG_TEXT_EXT = /* @__PURE__ */ new Set([
   ".sql",
   ".har"
 ]);
+var VISUAL_EXT = /* @__PURE__ */ new Set([".png", ".jpg", ".jpeg", ".gif", ".webp", ".bmp", ".pdf"]);
 function withoutHeredocs(cmd) {
   return cmd.replace(/<<-?\s*(['"]?)(\w+)\1[^\n]*\n[\s\S]*?\n\s*\2\s*(?=\n|$)/g, "<<$2");
 }
@@ -31774,10 +31891,11 @@ function suggest(toolName, input2, root, extra = {}) {
     case "Read": {
       const fp = String(input2.file_path ?? "");
       if (!fp) return void 0;
+      const ext = extname2(fp).toLowerCase();
+      if (VISUAL_EXT.has(ext)) return void 0;
       try {
         const abs = resolve4(root, fp);
         const st = statSync5(abs);
-        const ext = extname2(fp).toLowerCase();
         if (st.size > 200 * 1024 || BIG_TEXT_EXT.has(ext) && st.size > 20 * 1024) {
           if (!checkReadPath(loadPermissionRules(root), abs).allowed) return void 0;
           return {
@@ -32193,10 +32311,12 @@ ${guide}`);
       parts.push(
         "No onboarding memories yet: run /eminiar:em-onboard once so future sessions start with the project's structure, build and test commands."
       );
-    else if (!readManifest(ctx).sources.length)
+    else if (!readManifest(ctx).sources.length) {
+      const orphaned = db.documentSourceStats(DOCS_PREFIX).length;
       parts.push(
-        `No documentation sources registered yet: em_docs {action:"discover"} maps this project's libraries, services and APIs to their documentation, and {action:"add", pick:"recommended"} indexes it for em_docs lookup.`
+        orphaned ? `No documentation sources registered (${orphaned} orphaned docs:* source(s) still in the index; em_docs {action:"list"} shows them, {action:"remove"} drops them, {action:"add", pick:"recommended"} re-registers).` : `No documentation sources registered yet: em_docs {action:"discover"} maps this project's libraries, services and APIs to their documentation, and {action:"add", pick:"recommended"} indexes it for em_docs lookup.`
       );
+    }
   } catch {
   }
   try {