opencode-sesh 2.1.5 → 2.1.6

package/AGENTS.md

@@ -198,7 +198,10 @@ when opencode exits, the picker reopens with the query preserved.
   at `SESSION_MAX` and reported when truncated) rather than a fixed-size
   window, then indexes transcript text for **every** session in batches
   (`TRANSCRIPT_BATCH`) with bounded remote concurrency (`REMOTE_CONCURRENCY`).
-  The header shows indexing progress/coverage. Visible transcript hits get
+  Local extraction is one SQL statement per batch in the DB worker
+  (`transcriptTextQuery`, JSON-guarded) that returns only `{sid, text}` rows,
+  so tool payloads never reach the UI thread; if the worker rejects, sessions
+  not yet indexed fall back to the remote path. The header shows indexing progress/coverage. Visible transcript hits get
   short excerpts from the text-only index (never tool or reasoning payloads).
   Project, needs-input (Option-W), and pinned-session (Option-S) filters compose
   with the query; the picker remembers query/scope/filters until plugin restart.
@@ -273,7 +276,8 @@ when opencode exits, the picker reopens with the query preserved.
   (also run by `bun run test`).
 - `bun run test:tui-worker` — starts the real blob Worker against a bun:sqlite
   fixture; asserts results round-trip and the main thread keeps ticking during
-  a slow query (also run by `bun run test`).
+  a slow query and during transcript indexing of large tool payloads (also run
+  by `bun run test`).
 - `bun run typecheck` — `tsc` over `tui-plugins/` and `plugins/`.
 - `bun run lint:sh` — `bash -n` on every script.
 - `HOME=/tmp/fakehome bash install.sh` — installer smoke test.

package/CHANGELOG.md

@@ -5,6 +5,12 @@ All notable changes to this project are documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [2.1.6] - 2026-11-02
+
+### Changed
+
+- **tui:** Extract transcript text in the DB worker (#35)
+
 ## [2.1.5] - 2026-11-02
 
 ### Fixed
@@ -542,6 +548,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Fixture-database regression suite, TypeScript typechecking, and ShellCheck in
   CI.
 
+[2.1.6]: https://github.com/ozandogrultan/opencode-sesh/compare/v2.1.5...v2.1.6
 [2.1.5]: https://github.com/ozandogrultan/opencode-sesh/compare/v2.1.4...v2.1.5
 [2.1.4]: https://github.com/ozandogrultan/opencode-sesh/compare/v2.1.3...v2.1.4
 [2.1.3]: https://github.com/ozandogrultan/opencode-sesh/compare/v2.1.2...v2.1.3

package/dist/tui.js

@@ -920,29 +920,39 @@ GROUP BY live.session_id`, batch);
   }
   return details;
 }
-function addTranscript(index, sid, data) {
-  try {
-    const message = JSON.parse(data);
-    const texts = [];
-    if (typeof message?.text === "string") texts.push(message.text);
-    if (Array.isArray(message?.content)) {
-      for (const part of message.content) {
-        if (part?.type === "text" && typeof part.text === "string") texts.push(part.text);
-      }
-    }
-    for (const text of texts) {
-      const base = index.get(sid) ?? "";
-      index.set(sid, `${base} ${text}`.trim());
-    }
-  } catch {
-    // ignore malformed messages
-  }
+
+// Text parts only, extracted inside SQLite so tool and reasoning payloads never
+// reach the caller. The CASE guards keep json_extract away from malformed rows
+// and non-object parts, which SQL AND ordering alone would not.
+function transcriptTextQuery(count) {
+  const placeholders = Array.from({
+    length: count
+  }, () => "?").join(",");
+  return `WITH msg AS (
+  SELECT session_id AS sid, seq, rowid AS rid, CASE WHEN json_valid(data) THEN data END AS doc
+  FROM session_message
+  WHERE session_id IN (${placeholders}) AND type IN ('user', 'assistant')
+), parts AS (
+  SELECT sid, seq, rid, 0 AS part,
+    CASE WHEN json_type(doc, '$.text') = 'text' THEN json_extract(doc, '$.text') END AS text
+  FROM msg
+  UNION ALL
+  SELECT msg.sid, msg.seq, msg.rid, 1 + item.key AS part,
+    CASE WHEN item.type = 'object' THEN
+      CASE WHEN json_extract(item.value, '$.type') = 'text' AND json_type(item.value, '$.text') = 'text'
+        THEN json_extract(item.value, '$.text') END
+    END AS text
+  FROM msg, json_each(msg.doc, '$.content') AS item
+  WHERE json_type(msg.doc, '$.content') = 'array'
+)
+SELECT sid, text FROM parts WHERE text IS NOT NULL AND text <> '' ORDER BY sid, seq, rid, part`;
 }
 
 // Index titles/directories immediately, then add transcript text progressively
 // in batches so the render thread can breathe. Covers every entry, not a fixed
-// newest window; `onProgress` lets the picker show coverage.
-async function buildSearchIndex(api, entries, onProgress, active = () => true) {
+// newest window; `onProgress` lets the picker show coverage. Local extraction
+// runs through `run` (the worker client) and returns text only.
+async function buildSearchIndex(api, entries, onProgress, active = () => true, run = dbClient.run) {
   const index = new Map();
   for (const entry of entries) {
     index.set(entry.id, `${entry.title} ${entry.dir}`);
@@ -960,25 +970,25 @@ async function buildSearchIndex(api, entries, onProgress, active = () => true) {
   };
   report(false);
   if (!active()) return index;
-  const db = await openTranscriptDb();
-  if (db) {
-    try {
-      for (let i = 0; i < ids.length; i += TRANSCRIPT_BATCH) {
-        if (!active()) break;
-        const batch = ids.slice(i, i + TRANSCRIPT_BATCH);
-        const placeholders = batch.map(() => "?").join(",");
-        const rows = db.query(`SELECT session_id AS sid, data FROM session_message WHERE session_id IN (${placeholders}) AND type IN ('user', 'assistant') ORDER BY seq`).all(...batch) ?? [];
-        for (const row of rows) addTranscript(index, row.sid, row.data);
-        for (const id of batch) remaining.delete(id);
-        indexed += batch.length;
-        report(false);
-        await nextTick();
+  try {
+    for (let i = 0; i < ids.length; i += TRANSCRIPT_BATCH) {
+      if (!active()) break;
+      const batch = ids.slice(i, i + TRANSCRIPT_BATCH);
+      const rows = await run(transcriptTextQuery(batch.length), batch);
+      if (!active()) break;
+      const texts = new Map();
+      for (const row of rows) {
+        const list = texts.get(row.sid);
+        if (list) list.push(row.text);else texts.set(row.sid, [row.text]);
       }
-    } catch {
-      // local store unavailable mid-walk; fall back to the remote path below
-    } finally {
-      db.close();
+      for (const [sid, list] of texts) index.set(sid, [index.get(sid) ?? "", ...list].join(" ").trim());
+      for (const id of batch) remaining.delete(id);
+      indexed += batch.length;
+      report(false);
+      await nextTick();
     }
+  } catch {
+    // local store unavailable mid-walk; fall back to the remote path below
   }
   if (active() && remaining.size > 0) {
     await buildSearchIndexRemote(api, [...remaining], index, () => {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "opencode-sesh",
-  "version": "2.1.5",
+  "version": "2.1.6",
   "description": "OpenCode session browser: search, preview, and resume every session across every project from the terminal or TUI.",
   "type": "module",
   "license": "MIT",

package/tui-plugins/sesh-panel/tui.tsx

@@ -805,33 +805,41 @@ GROUP BY live.session_id`, batch)) as { id: string; question_since: number | nul
   return details
 }
 
-function addTranscript(index: Map<string, string>, sid: string, data: string): void {
-  try {
-    const message = JSON.parse(data) as { text?: unknown; content?: unknown }
-    const texts: string[] = []
-    if (typeof message?.text === "string") texts.push(message.text)
-    if (Array.isArray(message?.content)) {
-      for (const part of message.content as { type?: unknown; text?: unknown }[]) {
-        if (part?.type === "text" && typeof part.text === "string") texts.push(part.text)
-      }
-    }
-    for (const text of texts) {
-      const base = index.get(sid) ?? ""
-      index.set(sid, `${base} ${text}`.trim())
-    }
-  } catch {
-    // ignore malformed messages
-  }
+// Text parts only, extracted inside SQLite so tool and reasoning payloads never
+// reach the caller. The CASE guards keep json_extract away from malformed rows
+// and non-object parts, which SQL AND ordering alone would not.
+function transcriptTextQuery(count: number): string {
+  const placeholders = Array.from({ length: count }, () => "?").join(",")
+  return `WITH msg AS (
+  SELECT session_id AS sid, seq, rowid AS rid, CASE WHEN json_valid(data) THEN data END AS doc
+  FROM session_message
+  WHERE session_id IN (${placeholders}) AND type IN ('user', 'assistant')
+), parts AS (
+  SELECT sid, seq, rid, 0 AS part,
+    CASE WHEN json_type(doc, '$.text') = 'text' THEN json_extract(doc, '$.text') END AS text
+  FROM msg
+  UNION ALL
+  SELECT msg.sid, msg.seq, msg.rid, 1 + item.key AS part,
+    CASE WHEN item.type = 'object' THEN
+      CASE WHEN json_extract(item.value, '$.type') = 'text' AND json_type(item.value, '$.text') = 'text'
+        THEN json_extract(item.value, '$.text') END
+    END AS text
+  FROM msg, json_each(msg.doc, '$.content') AS item
+  WHERE json_type(msg.doc, '$.content') = 'array'
+)
+SELECT sid, text FROM parts WHERE text IS NOT NULL AND text <> '' ORDER BY sid, seq, rid, part`
 }
 
 // Index titles/directories immediately, then add transcript text progressively
 // in batches so the render thread can breathe. Covers every entry, not a fixed
-// newest window; `onProgress` lets the picker show coverage.
+// newest window; `onProgress` lets the picker show coverage. Local extraction
+// runs through `run` (the worker client) and returns text only.
 async function buildSearchIndex(
   api: TuiPluginApi,
   entries: Entry[],
   onProgress?: (progress: IndexProgress, index: Map<string, string>) => void,
   active = () => true,
+  run: DbRun = dbClient.run,
 ): Promise<Map<string, string>> {
   const index = new Map<string, string>()
   for (const entry of entries) {
@@ -845,29 +853,26 @@ async function buildSearchIndex(
   report(false)
 
   if (!active()) return index
-  const db = await openTranscriptDb()
-  if (db) {
-    try {
-      for (let i = 0; i < ids.length; i += TRANSCRIPT_BATCH) {
-        if (!active()) break
-        const batch = ids.slice(i, i + TRANSCRIPT_BATCH)
-        const placeholders = batch.map(() => "?").join(",")
-        const rows = (db
-          .query(
-            `SELECT session_id AS sid, data FROM session_message WHERE session_id IN (${placeholders}) AND type IN ('user', 'assistant') ORDER BY seq`,
-          )
-          .all(...batch) ?? []) as { sid: string; data: string }[]
-        for (const row of rows) addTranscript(index, row.sid, row.data)
-        for (const id of batch) remaining.delete(id)
-        indexed += batch.length
-        report(false)
-        await nextTick()
+  try {
+    for (let i = 0; i < ids.length; i += TRANSCRIPT_BATCH) {
+      if (!active()) break
+      const batch = ids.slice(i, i + TRANSCRIPT_BATCH)
+      const rows = (await run(transcriptTextQuery(batch.length), batch)) as { sid: string; text: string }[]
+      if (!active()) break
+      const texts = new Map<string, string[]>()
+      for (const row of rows) {
+        const list = texts.get(row.sid)
+        if (list) list.push(row.text)
+        else texts.set(row.sid, [row.text])
       }
-    } catch {
-      // local store unavailable mid-walk; fall back to the remote path below
-    } finally {
-      db.close()
+      for (const [sid, list] of texts) index.set(sid, [index.get(sid) ?? "", ...list].join(" ").trim())
+      for (const id of batch) remaining.delete(id)
+      indexed += batch.length
+      report(false)
+      await nextTick()
     }
+  } catch {
+    // local store unavailable mid-walk; fall back to the remote path below
   }
 
   if (active() && remaining.size > 0) {