@loicngr/kobo 1.11.43 → 1.11.45

package/dist/server/services/agent/engines/codex/engine.js

@@ -124,14 +124,41 @@ export function createCodexEngine() {
             });
             void childFailurePromise.catch(() => { });
             const child = spawnAppServer({ cwd: options.workingDir, env: options.env, signal: abortController.signal });
+            let childExited = false;
+            let resolveChildExited;
+            const childExitedPromise = new Promise((resolve) => {
+                resolveChildExited = resolve;
+            });
+            const confirmChildExited = () => {
+                childExited = true;
+                resolveChildExited();
+            };
+            let shutdownPromise;
+            const terminateChild = () => {
+                if (childExited)
+                    return Promise.resolve();
+                shutdownPromise ??= (async () => {
+                    if (!(await signalAndWaitForExit(child, 'SIGTERM', CODEX_FORCE_KILL_TIMEOUT_MS))) {
+                        console.warn('[codex] app-server ignored SIGTERM — sending SIGKILL');
+                        if (!(await signalAndWaitForExit(child, 'SIGKILL', CODEX_FORCE_KILL_TIMEOUT_MS))) {
+                            throw new Error('Codex app-server did not exit after SIGKILL');
+                        }
+                    }
+                    confirmChildExited();
+                })();
+                return shutdownPromise;
+            };
             child.on('error', (error) => {
                 const isExpectedAbort = abortController.signal.aborted && (error.code === 'ABORT_ERR' || error.name === 'AbortError');
                 if (isExpectedAbort)
                     return;
+                if (child.pid === undefined)
+                    confirmChildExited();
                 console.error('[codex] child process error:', error);
                 rejectChildFailure(error);
             });
             child.once('exit', (code, signal) => {
+                confirmChildExited();
                 if (!iteratorRunning || abortController.signal.aborted)
                     return;
                 const detail = code === null ? `signal ${signal ?? 'unknown'}` : `code ${code}`;
@@ -154,6 +181,7 @@ export function createCodexEngine() {
             let rejectTurnDone;
             const activeSubagentThreads = new Map();
             let waitingForBackgroundSubagents = false;
+            let watchdogEnded = false;
             const turnDonePromise = new Promise((resolve, reject) => {
                 resolveTurnDone = resolve;
                 rejectTurnDone = reject;
@@ -169,6 +197,8 @@ export function createCodexEngine() {
             // deadline tracks the last activity, not the start of the wait.
             const armSubagentStallWatchdog = () => {
                 clearSubagentStallWatchdog();
+                if (!waitingForBackgroundSubagents || pendingByCallId.size > 0 || activeSubagentThreads.size === 0)
+                    return;
                 subagentStallTimer = setTimeout(() => {
                     subagentStallTimer = undefined;
                     console.warn(`[codex-engine] Background subagent thread(s) still tracked active ${CODEX_SUBAGENT_STALL_TIMEOUT_MS}ms after the turn completed — forcing session drain.`);
@@ -176,7 +206,13 @@ export function createCodexEngine() {
                     // Forced, unclean termination (a thread may still be alive
                     // server-side) — never report it as a normal completion, so
                     // auto-loop doesn't treat an orphaned run as forward progress.
-                    mapperState.sawErrorResult = true;
+                    watchdogEnded = true;
+                    safeEmit({
+                        kind: 'error',
+                        category: 'other',
+                        code: 'subagent_idle_timeout',
+                        message: 'Session force-ended: background subagents stopped reporting activity (watchdog).',
+                    });
                     if (waitingForBackgroundSubagents)
                         resolveTurnDone();
                 }, CODEX_SUBAGENT_STALL_TIMEOUT_MS);
@@ -209,8 +245,13 @@ export function createCodexEngine() {
             const turnLiveness = createTurnLiveness({
                 timeoutMs: CODEX_TURN_IDLE_TIMEOUT_MS,
                 onTimeout() {
-                    mapperState.sawErrorResult = true;
-                    safeEmit({ kind: 'error', category: 'other', message: 'Codex stopped reporting activity for this turn' });
+                    watchdogEnded = true;
+                    safeEmit({
+                        kind: 'error',
+                        category: 'other',
+                        code: 'stream_idle_timeout',
+                        message: 'Codex stopped reporting activity for this turn',
+                    });
                     rejectTurnDone(new CodexTurnTimeoutError());
                 },
             });
@@ -221,6 +262,21 @@ export function createCodexEngine() {
                 rejectTurnDone(err);
             });
             const fileChanges = new Map();
+            const isParentTurn = (threadId, turnId) => (!threadId || threadId === discoveredSessionId) && (!activeTurnId || !turnId || turnId === activeTurnId);
+            const otherTurnMappers = new Map();
+            const mapTurnEvents = (notification, map) => {
+                if (isParentTurn(notification.threadId, notification.turnId))
+                    return map(mapperState);
+                const key = JSON.stringify([notification.threadId, notification.turnId]);
+                let state = otherTurnMappers.get(key);
+                if (!state) {
+                    state = createMapperState();
+                    otherTurnMappers.set(key, state);
+                }
+                // Child output remains visible, but its failures/compaction/markers
+                // cannot control the parent session or its auto-loop lifecycle.
+                return map(state).filter((event) => event.kind !== 'error' && !event.kind.startsWith('session:'));
+            };
             const itemKey = (threadId, turnId, itemId) => JSON.stringify([threadId, turnId, itemId]);
             const client = createAppServerClient({
                 stdin: child.stdin,
@@ -228,6 +284,10 @@ export function createCodexEngine() {
                 clientInfo: { name: 'kobo', version: getPackageVersion() },
                 onNotification(method, params) {
                     turnLiveness.activity();
+                    const notification = params;
+                    if (notification?.threadId && activeSubagentThreads.has(notification.threadId)) {
+                        armSubagentStallWatchdog();
+                    }
                     // Ignored notifications — harmless bookkeeping by the server
                     if (method === 'mcpServer/startupStatus/updated') {
                         const status = (params ?? {});
@@ -257,7 +317,7 @@ export function createCodexEngine() {
                         const n = params;
                         if (n.item.type === 'fileChange')
                             fileChanges.set(itemKey(n.threadId, n.turnId, n.item.id), n.item);
-                        const events = handleItemStarted(n.item, mapperState);
+                        const events = mapTurnEvents(n, (state) => handleItemStarted(n.item, state));
                         for (const ev of events)
                             safeEmit(ev);
                         if (n.item.type === 'collabAgentToolCall' && n.item.tool === 'spawnAgent') {
@@ -278,7 +338,7 @@ export function createCodexEngine() {
                         const n = params;
                         if (n.item.type === 'fileChange')
                             fileChanges.set(itemKey(n.threadId, n.turnId, n.item.id), n.item);
-                        const events = handleItemCompleted(n.item, mapperState);
+                        const events = mapTurnEvents(n, (state) => handleItemCompleted(n.item, state));
                         for (const ev of events)
                             safeEmit(ev);
                         if (n.item.type === 'collabAgentToolCall') {
@@ -302,7 +362,7 @@ export function createCodexEngine() {
                     }
                     if (method === 'item/agentMessage/delta') {
                         const n = params;
-                        for (const ev of handleAgentMessageDelta(n, mapperState))
+                        for (const ev of mapTurnEvents(n, (state) => handleAgentMessageDelta(n, state)))
                             safeEmit(ev);
                         return;
                     }
@@ -313,9 +373,9 @@ export function createCodexEngine() {
                             if (threadId === n.threadId && turnId === n.turn.id)
                                 fileChanges.delete(key);
                         }
-                        for (const ev of handleTurnCompleted(n, mapperState))
-                            safeEmit(ev);
-                        if (!activeTurnId || !n.turn?.id || n.turn.id === activeTurnId) {
+                        if (isParentTurn(n.threadId, n.turn.id)) {
+                            for (const ev of handleTurnCompleted(n, mapperState))
+                                safeEmit(ev);
                             if (n.turn?.status === 'completed' && activeSubagentThreads.size > 0) {
                                 waitingForBackgroundSubagents = true;
                                 turnLiveness.pause();
@@ -349,6 +409,8 @@ export function createCodexEngine() {
                     }
                     if (method === 'error') {
                         const n = params;
+                        if (!isParentTurn(n.threadId, n.turnId))
+                            return;
                         if (n.willRetry)
                             return;
                         const msg = n?.error?.message ?? 'unknown error';
@@ -371,6 +433,7 @@ export function createCodexEngine() {
                         register(callId, pending) {
                             pendingByCallId.set(callId, pending);
                             turnLiveness.pause();
+                            clearSubagentStallWatchdog();
                         },
                         respondError: (reqId, code, message) => client.peer.respondError(reqId, code, message),
                         respond: (reqId, result) => client.peer.respond(reqId, result),
@@ -444,7 +507,9 @@ export function createCodexEngine() {
                         ? 'error'
                         : mapperState.sawTurnInterrupted
                             ? 'killed'
-                            : 'completed';
+                            : watchdogEnded
+                                ? 'watchdog'
+                                : 'completed';
                     safeEmit({
                         kind: 'session:ended',
                         reason,
@@ -465,7 +530,11 @@ export function createCodexEngine() {
                         safeEmit({ kind: 'session:ended', reason: 'killed', exitCode: null });
                     }
                     else if (error.name === 'CodexTurnTimeoutError') {
-                        safeEmit({ kind: 'session:ended', reason: 'error', exitCode: null });
+                        safeEmit({
+                            kind: 'session:ended',
+                            reason: mapperState.sawErrorResult ? 'error' : 'watchdog',
+                            exitCode: null,
+                        });
                     }
                     else if (QUOTA_PATTERN.test(message)) {
                         tryEmitQuota(mapperState, safeEmit, message);
@@ -497,18 +566,16 @@ export function createCodexEngine() {
                         }
                     }
                     fileChanges.clear();
+                    otherTurnMappers.clear();
                     pendingByCallId.clear();
                     pendingToolCalls.clear();
                     client.close();
-                    try {
-                        child.kill('SIGTERM');
-                    }
-                    catch {
-                        // best-effort
-                    }
+                    await terminateChild().catch((error) => console.error('[codex] runtime shutdown unconfirmed:', error));
                 }
             })();
             const engineProcess = {
+                ready: readyPromise,
+                closed: Promise.all([iteratorPromise, childExitedPromise]).then(() => { }),
                 get pid() {
                     return child.pid;
                 },
@@ -516,7 +583,7 @@ export function createCodexEngine() {
                     return discoveredSessionId;
                 },
                 isAlive() {
-                    return iteratorRunning;
+                    return iteratorRunning || !childExited;
                 },
                 sendMessage(text) {
                     const steer = async () => {
@@ -581,17 +648,7 @@ export function createCodexEngine() {
                     catch {
                         // swallow
                     }
-                    // Escalate. The SIGTERM in the `finally` below is a request, not a
-                    // guarantee: a wedged `codex app-server` that ignores it outlives the
-                    // stop, the archive, the delete and Kōbō's own shutdown, holding the
-                    // worktree and its share of the model quota, with nothing left in
-                    // this process tracking it.
-                    if (!(await signalAndWaitForExit(child, 'SIGTERM', CODEX_FORCE_KILL_TIMEOUT_MS))) {
-                        console.warn('[codex] app-server ignored SIGTERM — sending SIGKILL');
-                        if (!(await signalAndWaitForExit(child, 'SIGKILL', CODEX_FORCE_KILL_TIMEOUT_MS))) {
-                            throw new Error('Codex app-server did not exit after SIGKILL');
-                        }
-                    }
+                    await terminateChild();
                 },
                 resolvePendingUserInput(callId, response) {
                     const pending = pendingByCallId.get(callId);
@@ -601,8 +658,12 @@ export function createCodexEngine() {
                     // Only resume the idle-timeout clock once every outstanding
                     // approval has been answered — a sibling request may still be
                     // waiting on a human decision.
-                    if (pendingByCallId.size === 0)
-                        turnLiveness.resume();
+                    if (pendingByCallId.size === 0) {
+                        if (waitingForBackgroundSubagents)
+                            armSubagentStallWatchdog();
+                        else
+                            turnLiveness.resume();
+                    }
                     const result = buildResponseForResolve(pending, response);
                     client.peer.respond(pending.requestId, result);
                     return true;