@gotcos/glasses-server 6.40.1 → 6.40.2

package/CHANGELOG.md

@@ -1,3 +1,34 @@
+## 6.40.2
+
+A finished answer could leave the stream open forever.
+
+An Ollama turn with tools streamed its whole answer, persisted it, and then
+never sent the terminal `done` event. On the glasses the turn counted past
+1,100 seconds while the reply had actually completed in 166 and was already
+saved in Messages. Double-tap to cancel did nothing, because there was no live
+job to cancel.
+
+The cause is one line in the query route, and it predates the tool loop. The
+SSE callbacks are declared inside `const sid = await callModelStreaming(...)`,
+so while that call is running `sid` is in its temporal dead zone. `onStart`
+never noticed because its own `sid` PARAMETER shadows the outer binding.
+`onDone` has no such parameter: it read the outer `sid`, threw
+ReferenceError, and — critically — threw AFTER setting `done = true` and
+BEFORE writing the terminal event, so the failure could not even fall through
+to the error path. The socket stayed open with the work already committed.
+
+Only Ollama hit it. The child-process bridges resolve outside that window, so
+Claude and Codex kept working; the in-process Ollama loop awaits its own
+finalize before returning, which lands the callback squarely inside the dead
+zone. Both readers now use the session id captured from `onStart`.
+
+The existing SSE contract test could never have caught this: it fires
+callbacks on a macrotask after the router resolves, with a comment asserting
+that bridges "never" call back synchronously. That assumption is now marked
+as covering the child-process bridges only, and a new test pins the
+in-process ordering. Reverting the fix makes all three of its cases hang,
+which is the production symptom exactly.
+
 ## 6.40.1
 
 The local model can read your meetings and memories.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@gotcos/glasses-server",
-  "version": "6.40.1",
+  "version": "6.40.2",
   "description": "COS Glasses \u2014 self-hosted AI heads-up-display server for Even G2 smart glasses, powered by Claude Code, Codex, Cursor Agent CLI, or local Ollama",
   "type": "module",
   "bin": {

package/server/routes/query.ts

@@ -114,6 +114,22 @@ queryRouter.post('/query', async (req, res) => {
   res.write(': keepalive\n\n')
 
   let done = false
+  // The session id AS REPORTED BY THE BRIDGE.
+  //
+  // `const sid = await callModelStreaming(...)` below is not assigned until the
+  // whole call resolves, but these callbacks fire DURING that await — so any
+  // callback that reads `sid` reads a temporal-dead-zone binding and throws
+  // ReferenceError. `onStart` never noticed because its own `sid` PARAMETER
+  // shadows the outer const; `onDone` has no such parameter and did throw,
+  // after setting `done = true` and before writing the terminal event, which
+  // left the SSE socket open forever with the answer already persisted.
+  //
+  // Measured 2026-08-26: an Ollama tool turn streamed its full answer and never
+  // sent `done`; the glasses counted past 1,100 seconds on a turn that had
+  // finished in 166. Claude survived only because its bridge resolves outside
+  // the window — the same line was one scheduling change away from hanging
+  // every provider.
+  let streamSessionId: string | undefined = typeof sessionId === 'string' ? sessionId : undefined
   const abortController = new AbortController()
   res.on('close', () => {
     if (!done) abortController.abort()
@@ -123,6 +139,7 @@ queryRouter.post('/query', async (req, res) => {
   try {
     const sid = await callModelStreaming(resolvedQuery || '', sessionId, {
       onStart: (model, sid, cliSessionId, metadata) => {
+        streamSessionId = sid
         if (!done) {
           const payload = { model, sessionId: sid, cliSessionId, ...metadata }
           res.write(`event: start\ndata: ${JSON.stringify(payload)}\n\n`)
@@ -161,7 +178,7 @@ queryRouter.post('/query', async (req, res) => {
           // expiry or make maintenance proof outrun the pending write.
           if (resolvedAttachments.ids.length > 0) {
             await getMediaStore().associate(resolvedAttachments.ids, {
-              sessionId: sid,
+              sessionId: streamSessionId ?? '',
               ...(validGlobalMsgNum ? { globalMsgNum: validGlobalMsgNum } : {}),
               messageEra: activeMessageEra,
             }).catch((err) => console.error('[query] attachment association failed:', err))
@@ -171,7 +188,7 @@ queryRouter.post('/query', async (req, res) => {
             const attachments = mergeMediaAttachmentRefs(attachmentRefs, metadata?.outputAttachments)
             const { outputAttachments: _outputAttachments, ...runMetadata } = metadata ?? {}
             const payload = {
-              text: fullText, sessionId: sid, model, cliSessionId, ...runMetadata,
+              text: fullText, sessionId: streamSessionId, model, cliSessionId, ...runMetadata,
               ...(attachments.length > 0 ? { attachments } : {}),
             }
             res.write(`event: done\ndata: ${JSON.stringify(payload)}\n\n`)