@north-light/crouter 0.3.167 → 0.3.169

package/dist/pi-extensions/canvas-bash-valve.js

@@ -27,7 +27,7 @@
 import { spawn } from 'node:child_process';
 import { closeSync, existsSync, mkdirSync, openSync, readFileSync, readSync, rmSync, statSync, writeFileSync, } from 'node:fs';
 import { homedir } from 'node:os';
-import { bashJobPaths, formatBashElapsed, newBashJobId } from '../core/bash-jobs.js';
+import { backgroundBashJob, bashJobPaths, formatBashElapsed, newBashJobId } from '../core/bash-jobs.js';
 import { createBashToolDefinition } from '@earendil-works/pi-coding-agent';
 // ---------------------------------------------------------------------------
 // Deadline — 5 minutes by default, overridable via CRTR_BASH_VALVE_MS
@@ -49,7 +49,8 @@ function resolveValveMs() {
 //
 // Positional argv (never interpolated into the script — the command text
 // lives only in cmd.sh):
-//   $1 cmd.sh   $2 job.log   $3 job.exit   $4 job.bg   $5 jobId   $6 nodeId
+//   $1 cmd.sh     $2 job.log   $3 job.exit   $4 job.run
+//   $5 job.done   $6 job.bg    $7 jobId      $8 nodeId
 //
 // `--` before the paths is a dummy $0 (a standard `bash -c script -- args`
 // idiom) so the real payload lands at $1 onward.
@@ -58,13 +59,38 @@ const SUPERVISOR_SCRIPT = [
     'bash "$1" > "$2" 2>&1',
     'ec=$?',
     'echo "$ec" > "$3"',
-    'if [ -f "$4" ]; then',
-    '  crtr node message send "Background bash job $5 finished (exit $ec). Log: $2  Command: $1" --to "$6" --tier urgent || true',
+    'if mv "$4" "$5" 2>/dev/null; then',
+    '  :',
+    'elif [ -f "$6" ]; then',
+    '  crtr node message send "Background bash job $7 finished (exit $ec). Log: $2  Command: $1" --to "$8" --tier urgent || true',
     'fi',
 ].join('\n');
 function makeJobPaths(contextDir) {
     return bashJobPaths(contextDir, newBashJobId());
 }
+const liveForegroundJobs = new Set();
+let terminationHandoffInstalled = false;
+/** The broker is about to exit. Persist every foreground handoff synchronously
+ * before pi's own termination listener aborts its active tool calls. */
+function backgroundLiveJobsForTermination() {
+    for (const paths of [...liveForegroundJobs]) {
+        try {
+            backgroundBashJob(paths);
+            liveForegroundJobs.delete(paths);
+        }
+        catch (err) {
+            process.stderr.write(`crtr: failed to background bash job ${paths.jobId} during broker termination: ${err.message}\n`);
+        }
+    }
+}
+function installTerminationHandoff() {
+    if (terminationHandoffInstalled)
+        return;
+    terminationHandoffInstalled = true;
+    process.prependListener('SIGTERM', backgroundLiveJobsForTermination);
+    if (process.platform !== 'win32')
+        process.prependListener('SIGHUP', backgroundLiveJobsForTermination);
+}
 function killGroup(pgid) {
     try {
         process.kill(-pgid, 'SIGTERM');
@@ -194,6 +220,7 @@ export function createValveOperations(nodeId, contextDir) {
                 mkdirSync(paths.dir, { recursive: true });
                 writeFileSync(paths.cmdSh, command);
                 writeFileSync(paths.jobLog, '');
+                writeFileSync(paths.jobRun, '');
                 let offset = 0;
                 let settled = false;
                 let pgid;
@@ -205,6 +232,7 @@ export function createValveOperations(nodeId, contextDir) {
                         clearInterval(interval);
                     if (signal)
                         signal.removeEventListener('abort', onAbort);
+                    liveForegroundJobs.delete(paths);
                 };
                 const finish = (act) => {
                     if (settled)
@@ -213,7 +241,7 @@ export function createValveOperations(nodeId, contextDir) {
                     teardown();
                     act();
                 };
-                function background(auto = false) {
+                function finishBackground(auto = false) {
                     const elapsed = formatBashElapsed(Date.now() - startedAt);
                     const notice = `\n[backgrounded as job ${paths.jobId} after ${elapsed} in the foreground${auto ? ' (auto)' : ''}. ` +
                         `It will interrupt you with a message when it exits. ` +
@@ -222,8 +250,35 @@ export function createValveOperations(nodeId, contextDir) {
                     finish(() => resolve({ exitCode: 0 }));
                     // Backgrounded — job dir persists (inspectable); no cleanup here.
                 }
+                function finishForeground() {
+                    offset = tailOnce(paths.jobLog, offset, onData);
+                    const exitCode = readExitCode(paths.jobExit);
+                    finish(() => resolve({ exitCode }));
+                    cleanupJobDir(paths.dir);
+                }
+                function requestBackground(auto = false) {
+                    const state = backgroundBashJob(paths);
+                    if (state === 'backgrounded' || state === 'already-backgrounded') {
+                        finishBackground(auto);
+                        return;
+                    }
+                    if (state === 'finished') {
+                        finishForeground();
+                        return;
+                    }
+                    finish(() => reject(new Error(`bash job ${paths.jobId} disappeared before it could be backgrounded`)));
+                    cleanupJobDir(paths.dir);
+                }
                 function onAbort() {
-                    tailOnce(paths.jobLog, offset, onData);
+                    offset = tailOnce(paths.jobLog, offset, onData);
+                    if (existsSync(paths.jobBg)) {
+                        finishBackground();
+                        return;
+                    }
+                    // Once the command has written its exit code, let the supervisor
+                    // finish the job.run → job.done transition before cleanup.
+                    if (existsSync(paths.jobExit))
+                        return;
                     if (pgid)
                         killGroup(pgid);
                     finish(() => reject(new Error('aborted')));
@@ -234,7 +289,7 @@ export function createValveOperations(nodeId, contextDir) {
                 const execution = resolveExecutionCwd(cwd);
                 if (execution.warning)
                     onData(Buffer.from(execution.warning));
-                const child = spawn('bash', ['-c', SUPERVISOR_SCRIPT, '--', paths.cmdSh, paths.jobLog, paths.jobExit, paths.jobBg, paths.jobId, nodeId], {
+                const child = spawn('bash', ['-c', SUPERVISOR_SCRIPT, '--', paths.cmdSh, paths.jobLog, paths.jobExit, paths.jobRun, paths.jobDone, paths.jobBg, paths.jobId, nodeId], {
                     cwd: execution.cwd,
                     detached: true,
                     stdio: 'ignore',
@@ -242,27 +297,31 @@ export function createValveOperations(nodeId, contextDir) {
                 });
                 child.on('error', (err) => {
                     finish(() => reject(err));
+                    cleanupJobDir(paths.dir);
                 });
                 pgid = child.pid;
                 // The detached supervisor IS the group leader, so its pid is the pgid.
                 // Persisting it is what lets a human stop the job from the Inspector
                 // later; the agent's own handoff notice quotes the same number.
-                if (pgid !== undefined)
+                if (pgid !== undefined) {
                     writeFileSync(paths.jobPgid, String(pgid));
+                    liveForegroundJobs.add(paths);
+                }
                 child.unref();
                 interval = setInterval(() => {
                     offset = tailOnce(paths.jobLog, offset, onData);
-                    if (existsSync(paths.jobExit)) {
-                        offset = tailOnce(paths.jobLog, offset, onData);
-                        const exitCode = readExitCode(paths.jobExit);
-                        finish(() => resolve({ exitCode }));
-                        cleanupJobDir(paths.dir);
+                    if (existsSync(paths.jobDone)) {
+                        finishForeground();
                         return;
                     }
                     if (existsSync(paths.jobBg)) {
-                        background();
+                        finishBackground();
                         return;
                     }
+                    // job.exit is written before the supervisor claims job.done. Do not
+                    // time out or delete the directory during that in-progress exit.
+                    if (existsSync(paths.jobExit))
+                        return;
                     const elapsedMs = Date.now() - startedAt;
                     if (timeout !== undefined && timeout > 0) {
                         if (elapsedMs >= timeout * 1000) {
@@ -273,10 +332,8 @@ export function createValveOperations(nodeId, contextDir) {
                         }
                         return;
                     }
-                    if (elapsedMs >= valveMs) {
-                        writeFileSync(paths.jobBg, '');
-                        background(true);
-                    }
+                    if (elapsedMs >= valveMs)
+                        requestBackground(true);
                 }, POLL_MS);
             });
         },
@@ -292,6 +349,7 @@ export default function (pi) {
     const contextDir = process.env['CRTR_CONTEXT_DIR'];
     if (contextDir === undefined || contextDir.trim() === '')
         return; // defensive; always set alongside CRTR_NODE_ID
+    installTerminationHandoff();
     // Registered on session_start (fires for startup, new, resume, fork, and
     // /reload) so we get ctx.cwd — the same session cwd the builtin bash tool
     // is built against. registerTool replaces the builtin by name; re-firing on