@linxiraos/pi-utils 1.1.5 → 1.1.7

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/src/ptree.ts CHANGED
@@ -15,6 +15,83 @@ type InMask = "pipe" | "ignore" | Buffer | Uint8Array | null;
15
15
  /** A Bun subprocess with stdout/stderr always piped (stdin may vary). */
16
16
  type PipedSubprocess<In extends InMask = InMask> = Subprocess<In, "pipe", "pipe">;
17
17
 
18
+ const LINUX_SUBREAPER_COMMAND_ENV = "OMP_PTREE_SUBREAPER_COMMAND";
19
+ const LINUX_SUBREAPER_BUN_BE_BUN_ENV = "OMP_PTREE_SUBREAPER_BUN_BE_BUN";
20
+
21
+ /**
22
+ * Build the Linux child-subreaper entrypoint.
23
+ *
24
+ * @internal Exported so tests can force a missing first libc soname and verify
25
+ * the loader continues to the next candidate.
26
+ */
27
+ export function createLinuxSubreaperScript(libcCandidates: readonly string[] = ["libc.so.6", "libc.so"]): string {
28
+ return `
29
+ import { dlopen, FFIType } from "bun:ffi";
30
+
31
+ let libc;
32
+ for (const soname of ${JSON.stringify(libcCandidates)}) {
33
+ try {
34
+ libc = dlopen(soname, {
35
+ prctl: {
36
+ args: [FFIType.i32, FFIType.u64, FFIType.u64, FFIType.u64, FFIType.u64],
37
+ returns: FFIType.i32,
38
+ },
39
+ waitpid: {
40
+ args: [FFIType.i32, FFIType.ptr, FFIType.i32],
41
+ returns: FFIType.i32,
42
+ },
43
+ });
44
+ break;
45
+ } catch {}
46
+ }
47
+ if (!libc) throw new Error("failed to load libc for Linux child supervision");
48
+
49
+ if (libc.symbols.prctl(36, 1, 0, 0, 0) !== 0) {
50
+ throw new Error("failed to become a Linux child subreaper");
51
+ }
52
+
53
+ const commandJson = Bun.env.${LINUX_SUBREAPER_COMMAND_ENV};
54
+ if (!commandJson) throw new Error("missing supervised command");
55
+ const callerBunBeBun = Bun.env.${LINUX_SUBREAPER_BUN_BE_BUN_ENV};
56
+ delete Bun.env.${LINUX_SUBREAPER_COMMAND_ENV};
57
+ delete Bun.env.${LINUX_SUBREAPER_BUN_BE_BUN_ENV};
58
+ if (callerBunBeBun === undefined) delete Bun.env.BUN_BE_BUN;
59
+ else Bun.env.BUN_BE_BUN = callerBunBeBun;
60
+ const command = JSON.parse(commandJson);
61
+ const child = Bun.spawn(command, {
62
+ stdin: "inherit",
63
+ stdout: "pipe",
64
+ stderr: "pipe",
65
+ windowsHide: true,
66
+ env: Bun.env,
67
+ });
68
+
69
+ async function relay(stream, destination) {
70
+ const writer = destination.writer();
71
+ for await (const chunk of stream) writer.write(chunk);
72
+ await writer.flush();
73
+ }
74
+
75
+ function hasLiveChildren() {
76
+ let childPid;
77
+ do {
78
+ childPid = libc.symbols.waitpid(-1, null, 1);
79
+ } while (childPid > 0);
80
+ return childPid === 0;
81
+ }
82
+
83
+ const [exitCode] = await Promise.all([
84
+ child.exited,
85
+ relay(child.stdout, Bun.stdout),
86
+ relay(child.stderr, Bun.stderr),
87
+ ]);
88
+ while (hasLiveChildren()) await Bun.sleep(10);
89
+ process.exit(exitCode ?? 1);
90
+ `;
91
+ }
92
+
93
+ const LINUX_SUBREAPER_SCRIPT = createLinuxSubreaperScript();
94
+
18
95
  // ── Exceptions ───────────────────────────────────────────────────────────────
19
96
 
20
97
  /**
@@ -103,13 +180,30 @@ export class ChildProcess<In extends InMask = InMask> {
103
180
  #exitReasonPending?: Exception;
104
181
  #stderrDone: Promise<void>;
105
182
  #exited: Promise<number>;
183
+ #openPipeReaders = 1;
184
+ // Pipe reads race this cutoff only when attachTimeout() configures a
185
+ // command deadline. Untimed commands preserve complete EOF-based capture.
186
+ #drainCutoff: Promise<void>;
187
+ #resolveDrainCutoff: () => void;
188
+ #timeoutTimer?: NodeJS.Timeout;
106
189
  #stderrStream?: ReadableStream<Uint8Array>;
107
-
190
+ // Termination in flight after kill(); aborted exits await it before reporting.
191
+ #terminating?: Promise<boolean | void>;
192
+ #terminateGroup: boolean;
193
+ #hardKillTree: boolean;
194
+ // Windows has no process groups. Retaining the root's native handle pins
195
+ // its PID after exit so killTree() can still enumerate its original children.
196
+ #windowsRootProcess?: Process;
108
197
  constructor(
109
198
  readonly proc: PipedSubprocess<In>,
110
199
  readonly exposeStderr: boolean,
111
200
  retainFullStderr = exposeStderr,
201
+ terminateGroup = false,
202
+ hardKillTree = false,
112
203
  ) {
204
+ this.#terminateGroup = terminateGroup;
205
+ this.#hardKillTree = hardKillTree;
206
+ this.#windowsRootProcess = process.platform === "win32" ? (Process.fromPid(proc.pid) ?? undefined) : undefined;
113
207
  if (retainFullStderr) this.#stderrChunks = [];
114
208
  // Eagerly drain stderr into a truncated tail, retaining raw chunks only for explicit full capture.
115
209
  const dec = new TextDecoder();
@@ -123,22 +217,39 @@ export class ChildProcess<In extends InMask = InMask> {
123
217
  this.#stderrStream = teeStream;
124
218
  stderrStream = drainStream;
125
219
  }
220
+ // Normalize Bun's exited promise into our exitReason / exitedCleanly model.
221
+ const { promise, resolve, reject } = Promise.withResolvers<number>();
222
+ this.#exited = promise;
223
+ const drainCutoff = Promise.withResolvers<void>();
224
+ this.#drainCutoff = drainCutoff.promise;
225
+ this.#resolveDrainCutoff = drainCutoff.resolve;
226
+ // The cutoff remains pending for untimed commands, preserving complete
227
+ // EOF-based capture. attachTimeout() resolves it at the command deadline.
228
+
229
+ const pipeCutoff = this.#drainCutoff;
126
230
  this.#stderrDone = (async () => {
231
+ const reader = stderrStream.getReader();
127
232
  try {
128
- for await (const chunk of stderrStream) {
129
- this.#stderrChunks?.push(chunk);
130
- this.#stderrTail += dec.decode(chunk, { stream: true });
233
+ for (;;) {
234
+ const chunk = await Promise.race([
235
+ reader.read().then(r => ({ cutoff: false as const, r })),
236
+ pipeCutoff.then(() => ({ cutoff: true as const })),
237
+ ]);
238
+ if (chunk.cutoff) {
239
+ await reader.cancel().catch(() => {});
240
+ break;
241
+ }
242
+ if (chunk.r.done) break;
243
+ this.#stderrChunks?.push(chunk.r.value);
244
+ this.#stderrTail += dec.decode(chunk.r.value, { stream: true });
131
245
  trim();
132
246
  }
133
247
  } catch {}
248
+ this.#openPipeReaders--;
134
249
  this.#stderrTail += dec.decode();
135
250
  trim();
136
251
  })();
137
252
 
138
- // Normalize Bun's exited promise into our exitReason / exitedCleanly model.
139
- const { promise, resolve, reject } = Promise.withResolvers<number>();
140
- this.#exited = promise;
141
-
142
253
  proc.exited
143
254
  .catch(() => null)
144
255
  .then(async exitCode => {
@@ -153,6 +264,11 @@ export class ChildProcess<In extends InMask = InMask> {
153
264
  }
154
265
 
155
266
  await this.#stderrDone;
267
+ if (this.#exitReasonPending) {
268
+ this.#exitReason = this.#exitReasonPending;
269
+ reject(this.#exitReasonPending);
270
+ return;
271
+ }
156
272
 
157
273
  if (exitCode !== null) {
158
274
  this.#exitReason = new NonZeroExitError(exitCode, this.#stderrTail);
@@ -218,44 +334,160 @@ export class ChildProcess<In extends InMask = InMask> {
218
334
  }
219
335
 
220
336
  kill(reason?: Exception, gracefulMs?: number) {
221
- if (reason && !this.#exitReasonPending) this.#exitReasonPending = reason;
222
- if (!this.proc.killed)
223
- void Process.fromPid(this.proc.pid)
224
- ?.terminate(gracefulMs === undefined ? undefined : { gracefulMs })
337
+ if (reason && !this.#exitReasonPending) {
338
+ this.#exitReasonPending = reason;
339
+ // The normalized exit promise may already have resolved from a dead
340
+ // group leader; wait() still needs to report the later deadline.
341
+ if (this.proc.exitCode !== null) this.#exitReason = reason;
342
+ }
343
+ if (gracefulMs !== undefined && gracefulMs < 0 && this.#hardKillTree && this.proc.exitCode === null) {
344
+ // terminate() sends its polite wave to the root before rebuilding the
345
+ // hard-kill tree. A subreaper root can die in that gap and release its
346
+ // adopted descendants, so snapshot and hard-kill the live tree first.
347
+ const root = Process.fromPid(this.proc.pid);
348
+ if (root) {
349
+ root.killTree(9);
350
+ this.#terminating = Promise.resolve();
351
+ return;
352
+ }
353
+ }
354
+ if (
355
+ this.proc.exitCode !== null &&
356
+ this.#terminateGroup &&
357
+ this.#openPipeReaders > 0 &&
358
+ process.platform !== "win32"
359
+ ) {
360
+ // Bun detached children are POSIX session/process-group leaders. If
361
+ // the leader has exited, the native Process handle cannot rediscover
362
+ // its PGID, but a pipe-holding descendant keeps that exact group alive.
363
+ try {
364
+ process.kill(-this.proc.pid, "SIGKILL");
365
+ } catch {}
366
+ this.#terminating = Promise.resolve();
367
+ return;
368
+ }
369
+ if (this.proc.exitCode !== null && this.#windowsRootProcess && this.#openPipeReaders > 0) {
370
+ // The retained handle keeps the dead root PID reserved, making the
371
+ // Windows Toolhelp descendant walk identity-safe after root exit.
372
+ this.#windowsRootProcess.killTree();
373
+ this.#terminating = Promise.resolve();
374
+ return;
375
+ }
376
+ if (!this.proc.killed) {
377
+ const options =
378
+ gracefulMs === undefined
379
+ ? this.#terminateGroup
380
+ ? { group: true }
381
+ : undefined
382
+ : { gracefulMs, group: this.#terminateGroup };
383
+ this.#terminating = (this.#windowsRootProcess ?? Process.fromPid(this.proc.pid))
384
+ ?.terminate(options)
225
385
  ?.catch(e => void e);
386
+ }
226
387
  }
227
388
 
228
389
  // ── Output helpers ───────────────────────────────────────────────────
229
390
 
391
+ async #throwIfAborted(): Promise<void> {
392
+ const exitReason = this.exitReason;
393
+ if (!exitReason?.aborted) return;
394
+ if (this.#terminating) await this.#terminating;
395
+ throw exitReason;
396
+ }
397
+
230
398
  async text(): Promise<string> {
231
- const p = new Response(this.stdout).text();
399
+ const p = this.#readStream(this.proc.stdout);
232
400
  if (this.#nothrow) return p;
233
401
  const [text] = await Promise.all([p, this.exitedCleanly]);
402
+ await this.#throwIfAborted();
234
403
  return text;
235
404
  }
236
405
 
237
- async blob(): Promise<Blob> {
238
- const p = new Response(this.stdout).blob();
406
+ /**
407
+ * Read a pipe fully, stopping early only at an explicit command deadline.
408
+ */
409
+ async #readStream(stream: ReadableStream<Uint8Array>): Promise<string> {
410
+ this.#openPipeReaders++;
411
+ const reader = stream.getReader();
412
+ const dec = new TextDecoder();
413
+ let out = "";
414
+ try {
415
+ for (;;) {
416
+ const chunk = await Promise.race([
417
+ reader.read().then(r => ({ cutoff: false as const, r })),
418
+ this.#drainCutoff.then(() => ({ cutoff: true as const })),
419
+ ]);
420
+ if (chunk.cutoff) {
421
+ await reader.cancel().catch(() => {});
422
+ break;
423
+ }
424
+ if (chunk.r.done) break;
425
+ out += dec.decode(chunk.r.value, { stream: true });
426
+ }
427
+ } catch {
428
+ // A cancelled or failed read keeps whatever was already collected.
429
+ }
430
+ this.#openPipeReaders--;
431
+ return out + dec.decode();
432
+ }
433
+
434
+ async #readBytes(): Promise<Uint8Array> {
435
+ const reader = this.proc.stdout.getReader();
436
+ this.#openPipeReaders++;
437
+ const chunks: Uint8Array[] = [];
438
+ let length = 0;
439
+ try {
440
+ for (;;) {
441
+ const chunk = await Promise.race([
442
+ reader.read().then(r => ({ cutoff: false as const, r })),
443
+ this.#drainCutoff.then(() => ({ cutoff: true as const })),
444
+ ]);
445
+ if (chunk.cutoff) {
446
+ await reader.cancel().catch(() => {});
447
+ break;
448
+ }
449
+ if (chunk.r.done) break;
450
+ chunks.push(chunk.r.value);
451
+ length += chunk.r.value.byteLength;
452
+ }
453
+ } catch {
454
+ // A cancelled or failed read keeps whatever was already collected.
455
+ } finally {
456
+ this.#openPipeReaders--;
457
+ reader.releaseLock();
458
+ }
459
+
460
+ const bytes = new Uint8Array(length);
461
+ let offset = 0;
462
+ for (const chunk of chunks) {
463
+ bytes.set(chunk, offset);
464
+ offset += chunk.byteLength;
465
+ }
466
+ return bytes;
467
+ }
468
+
469
+ async #readOutputBytes(waitForCleanExit = false): Promise<Uint8Array> {
470
+ const p = this.#readBytes();
239
471
  if (this.#nothrow) return p;
240
- const [blob] = await Promise.all([p, this.exitedCleanly]);
241
- return blob;
472
+ const bytes = waitForCleanExit ? (await Promise.all([p, this.exitedCleanly]))[0] : await p;
473
+ await this.#throwIfAborted();
474
+ return bytes;
475
+ }
476
+
477
+ async blob(): Promise<Blob> {
478
+ return new Blob([await this.#readOutputBytes(true)]);
242
479
  }
243
480
 
244
481
  async json(): Promise<unknown> {
245
- return new Response(this.stdout).json();
482
+ return JSON.parse(new TextDecoder().decode(await this.#readOutputBytes()));
246
483
  }
247
484
 
248
485
  async arrayBuffer(): Promise<ArrayBuffer> {
249
- return new Response(this.stdout).arrayBuffer();
486
+ return (await this.#readOutputBytes()).buffer as ArrayBuffer;
250
487
  }
251
488
 
252
489
  async bytes(): Promise<Uint8Array> {
253
- // Bun's `Response(stream).bytes()` returns the raw `ArrayBuffer` once the
254
- // stream emits more than one chunk (subprocess stdout chunks past ~128 KB).
255
- // Normalize at the contract boundary so every caller — SSH read,
256
- // `decodeUtf8Text`, callers slicing with `.subarray` — sees a `Uint8Array`.
257
- const body = (await new Response(this.stdout).bytes()) as Uint8Array | ArrayBuffer;
258
- return body instanceof Uint8Array ? body : new Uint8Array(body);
490
+ return this.#readOutputBytes();
259
491
  }
260
492
 
261
493
  // ── Wait ─────────────────────────────────────────────────────────────
@@ -267,7 +499,7 @@ export class ChildProcess<In extends InMask = InMask> {
267
499
  throw new Error('Full stderr capture must be requested when spawning the process (pass stderr: "full")');
268
500
  }
269
501
 
270
- const stdoutP = new Response(this.stdout).text();
502
+ const stdoutP = this.#readStream(this.proc.stdout);
271
503
  const stderrP =
272
504
  stderrMode === "full" && stderrChunks
273
505
  ? this.#stderrDone.then(() => new TextDecoder().decode(Buffer.concat(stderrChunks)))
@@ -282,12 +514,17 @@ export class ChildProcess<In extends InMask = InMask> {
282
514
  if (err instanceof Exception) exitError = err;
283
515
  else throw err;
284
516
  }
285
-
517
+ this.#clearTimeout();
286
518
  if (!exitError) exitError = this.exitReason;
287
519
  if (!exitError && this.exitCode !== null && this.exitCode !== 0) {
288
520
  exitError = new NonZeroExitError(this.exitCode, this.#stderrTail);
289
521
  }
290
522
 
523
+ // On abort/timeout, hold the result until the tree is actually gone: the
524
+ // native terminate() is graceful-first, and reporting before it finishes
525
+ // would leave timed-out descendants alive past the caller's budget.
526
+ if (exitError?.aborted && this.#terminating) await this.#terminating;
527
+
291
528
  const exitCode = this.exitCode ?? (exitError && !exitError.aborted ? exitError.exitCode : null);
292
529
  const ok = exitCode === 0;
293
530
 
@@ -307,18 +544,32 @@ export class ChildProcess<In extends InMask = InMask> {
307
544
  this.#exited.catch(() => {}).finally(() => signal.removeEventListener("abort", onAbort));
308
545
  }
309
546
 
547
+ #clearTimeout(): void {
548
+ if (!this.#timeoutTimer) return;
549
+ clearTimeout(this.#timeoutTimer);
550
+ this.#timeoutTimer = undefined;
551
+ }
552
+
310
553
  attachTimeout(ms: number): void {
311
554
  if (ms <= 0 || this.proc.killed) return;
312
555
  this.#exited.catch(() => {});
313
- Promise.race([
314
- Bun.sleep(ms).then(() => true),
315
- this.proc.exited.then(
316
- () => false,
317
- () => false,
318
- ),
319
- ]).then(timedOut => {
320
- if (timedOut) this.kill(new TimeoutError(ms, this.#stderrTail));
321
- });
556
+ // One unref'd deadline controls both termination and pipe collection.
557
+ // A clean command clears it in wait(), so fast invocations do not hold
558
+ // the event loop for the unused remainder.
559
+ const timer = setTimeout(() => {
560
+ // A detached group can remain alive after its leader exits. Only use
561
+ // the dead-leader fallback while an inherited pipe proves that exact
562
+ // group still has a live member; this avoids stale-PGID reuse.
563
+ if (
564
+ this.proc.exitCode === null ||
565
+ (this.#openPipeReaders > 0 && (this.#terminateGroup || this.#windowsRootProcess))
566
+ ) {
567
+ this.kill(new TimeoutError(ms, this.#stderrTail), -1);
568
+ }
569
+ this.#resolveDrainCutoff();
570
+ }, ms);
571
+ timer.unref?.();
572
+ this.#timeoutTimer = timer;
322
573
  }
323
574
 
324
575
  [Symbol.dispose](): void {
@@ -336,6 +587,13 @@ type ChildSpawnOptions<In extends InMask = InMask> = Omit<
336
587
  > & {
337
588
  signal?: AbortSignal;
338
589
  detached?: boolean;
590
+ /**
591
+ * On Linux, supervise the command from a child subreaper so descendants
592
+ * remain reachable after changing session and reparenting. Other platforms
593
+ * ignore this option. macOS process groups cannot retain a daemonized
594
+ * descendant that creates a new session and reparents to launchd.
595
+ */
596
+ subreaper?: boolean;
339
597
  /** Expose and retain complete stderr for a later `wait({ stderr: "full" })`. */
340
598
  stderr?: "full" | null;
341
599
  };
@@ -345,15 +603,26 @@ function spawnInternal<In extends InMask = InMask>(
345
603
  opts: ChildSpawnOptions<In> | undefined,
346
604
  retainFullStderr: boolean,
347
605
  ): ChildProcess<In> {
348
- const { timeout = -1, signal, stderr, ...rest } = opts ?? {};
349
- const child = Bun.spawn(cmd, {
606
+ const { timeout = -1, signal, stderr, detached, subreaper = false, ...rest } = opts ?? {};
607
+ const useSubreaper = subreaper && process.platform === "linux";
608
+ const commandEnv = rest.env ?? Bun.env;
609
+ const child = Bun.spawn(useSubreaper ? [process.execPath, "-e", LINUX_SUBREAPER_SCRIPT] : cmd, {
350
610
  stdin: "ignore",
351
611
  stdout: "pipe",
352
612
  stderr: "pipe",
353
613
  windowsHide: true,
614
+ detached,
354
615
  ...rest,
616
+ env: useSubreaper
617
+ ? {
618
+ ...commandEnv,
619
+ BUN_BE_BUN: "1",
620
+ [LINUX_SUBREAPER_COMMAND_ENV]: JSON.stringify(cmd),
621
+ [LINUX_SUBREAPER_BUN_BE_BUN_ENV]: commandEnv.BUN_BE_BUN,
622
+ }
623
+ : rest.env,
355
624
  });
356
- const cp = new ChildProcess(child, stderr === "full", retainFullStderr);
625
+ const cp = new ChildProcess(child, stderr === "full", retainFullStderr, detached === true, useSubreaper);
357
626
  if (signal) cp.attachSignal(signal);
358
627
  if (timeout > 0) cp.attachTimeout(timeout);
359
628
  return cp;
@@ -2,6 +2,8 @@ import * as fs from "node:fs";
2
2
  import * as fsp from "node:fs/promises";
3
3
  import * as Module from "node:module";
4
4
  import * as path from "node:path";
5
+ import { withFileLock } from "./file-lock";
6
+ import { isEexist, isEnoent } from "./fs-error";
5
7
 
6
8
  /**
7
9
  * On-demand runtime dependency support for native-heavy optional packages
@@ -303,25 +305,62 @@ export interface EnsureRuntimeInstalledOptions {
303
305
  lockSleepMs?: number;
304
306
  }
305
307
 
306
- function isErrnoCode(error: unknown, code: string): boolean {
307
- return typeof error === "object" && error !== null && "code" in error && error.code === code;
308
- }
308
+ /** No runtime install plausibly runs this long, so older legacy lock directories are crash orphans. */
309
+ const STALE_LEGACY_LOCK_MS = 10 * 60_000;
309
310
 
310
- async function acquireInstallLock(runtimeDir: string, attempts: number, sleepMs: number): Promise<() => Promise<void>> {
311
- const lockDir = `${runtimeDir}.lock`;
312
- await fsp.mkdir(path.dirname(lockDir), { recursive: true });
313
- for (let attempt = 0; attempt < attempts; attempt++) {
311
+ /**
312
+ * Run `fn` while reserving the pre-crash-safe `${runtimeDir}.lock` namespace.
313
+ *
314
+ * Versions through 18.0.10 serialized installs with a bare lock *directory*
315
+ * that only its creator removed; an installer killed outside that window
316
+ * (SIGKILL/OOM/Ctrl-C) left it unreleasable, wedging every later install for
317
+ * the full wait envelope (issue #10120). During an in-flight upgrade a legacy
318
+ * process may still legitimately own this directory, so poll until it is
319
+ * released and only force-reclaim once the directory is older than any
320
+ * plausible install ({@link STALE_LEGACY_LOCK_MS}) — never merely because a
321
+ * retry budget elapsed, which would delete a still-active legacy lock and let
322
+ * two installers race the same tree. Once the namespace is free, atomically
323
+ * create and retain a regular file through `fn`: an older process cannot
324
+ * acquire it between the handoff check and the new install. A file left by a
325
+ * crashed new installer can be reused immediately because the outer OS lock
326
+ * proves its owner is gone, unlike a legacy directory whose owner is unknown.
327
+ */
328
+ async function withLegacyInstallLock<T>(runtimeDir: string, sleepMs: number, fn: () => Promise<T>): Promise<T> {
329
+ const legacy = `${runtimeDir}.lock`;
330
+ for (;;) {
314
331
  try {
315
- await fsp.mkdir(lockDir);
316
- return async () => {
317
- await fsp.rm(lockDir, { recursive: true, force: true });
318
- };
332
+ const reservation = await fsp.open(legacy, "wx");
333
+ await reservation.close();
319
334
  } catch (error) {
320
- if (!isErrnoCode(error, "EEXIST")) throw error;
321
- await Bun.sleep(sleepMs);
335
+ if (!isEexist(error)) throw error;
336
+ let stat: fs.Stats;
337
+ try {
338
+ stat = await fsp.stat(legacy);
339
+ } catch (statError) {
340
+ if (isEnoent(statError)) continue; // released between open and stat; retry
341
+ throw statError;
342
+ }
343
+ // A non-directory is a reservation left by a newer installer. The
344
+ // outer OS lock proves that installer is gone, so reuse it at once.
345
+ if (!stat.isDirectory()) break;
346
+ // A fresh directory may still belong to a live pre-18.x installer, so
347
+ // wait for it to finish; only a crash orphan (older than any plausible
348
+ // install) is force-reclaimed.
349
+ if (Date.now() - stat.mtimeMs > STALE_LEGACY_LOCK_MS) {
350
+ await fsp.rm(legacy, { recursive: true, force: true });
351
+ } else {
352
+ await Bun.sleep(sleepMs);
353
+ }
354
+ continue;
322
355
  }
356
+ break;
357
+ }
358
+ // Retain the regular-file reservation across the install.
359
+ try {
360
+ return await fn();
361
+ } finally {
362
+ await fsp.rm(legacy, { force: true });
323
363
  }
324
- throw new Error(`Timed out waiting for runtime install lock: ${lockDir}`);
325
364
  }
326
365
 
327
366
  export async function writeRuntimeManifest(runtimeDir: string, install: RuntimeInstallSpec): Promise<void> {
@@ -363,7 +402,14 @@ async function runRuntimeInstall(runtimeDir: string): Promise<void> {
363
402
 
364
403
  /**
365
404
  * Materialize a pinned dependency set into `runtimeDir` (idempotent,
366
- * cross-process safe via a lock directory). Returns `runtimeDir`.
405
+ * cross-process safe). Returns `runtimeDir`.
406
+ *
407
+ * Serialization uses the OS-backed {@link withFileLock} at
408
+ * `${runtimeDir}.install.lock`, which the kernel releases on process death, so
409
+ * a crashed installer cannot wedge later attempts (issue #10120). The path is
410
+ * deliberately distinct from the legacy `${runtimeDir}.lock` mkdir directory;
411
+ * {@link withLegacyInstallLock} atomically reserves that namespace during the
412
+ * new install so older processes cannot cross the migration boundary.
367
413
  */
368
414
  export async function ensureRuntimeInstalled(options: EnsureRuntimeInstalledOptions): Promise<string> {
369
415
  const { runtimeDir, install, onPhase, lockAttempts = 240, lockSleepMs = 250 } = options;
@@ -379,15 +425,20 @@ export async function ensureRuntimeInstalled(options: EnsureRuntimeInstalledOpti
379
425
  if (await probeManifest.exists()) return runtimeDir;
380
426
 
381
427
  onPhase?.("initiate");
382
- const releaseLock = await acquireInstallLock(runtimeDir, lockAttempts, lockSleepMs);
383
- try {
384
- if (await probeManifest.exists()) return runtimeDir;
385
- await writeRuntimeManifest(runtimeDir, install);
386
- onPhase?.("download");
387
- await runRuntimeInstall(runtimeDir);
388
- onPhase?.("done");
389
- return runtimeDir;
390
- } finally {
391
- await releaseLock();
392
- }
428
+ // withFileLock does not create parent directories; the runtime cache dir may
429
+ // not exist yet on the very first install.
430
+ await fsp.mkdir(path.dirname(runtimeDir), { recursive: true });
431
+ return withFileLock(
432
+ `${runtimeDir}.install`,
433
+ () =>
434
+ withLegacyInstallLock(runtimeDir, lockSleepMs, async () => {
435
+ if (await probeManifest.exists()) return runtimeDir;
436
+ await writeRuntimeManifest(runtimeDir, install);
437
+ onPhase?.("download");
438
+ await runRuntimeInstall(runtimeDir);
439
+ onPhase?.("done");
440
+ return runtimeDir;
441
+ }),
442
+ { retries: lockAttempts, retryDelayMs: lockSleepMs },
443
+ );
393
444
  }
package/src/sqlite.ts CHANGED
@@ -7,6 +7,12 @@
7
7
  * one implementation here prevents the classifiers from drifting between the
8
8
  * credential store and the model cache.
9
9
  */
10
+ import type { Database } from "bun:sqlite";
11
+
12
+ /** Checkpoints committed WAL frames without waiting for concurrent readers. */
13
+ export function checkpointWal(db: Database): void {
14
+ db.run("PRAGMA wal_checkpoint(PASSIVE)");
15
+ }
10
16
 
11
17
  /**
12
18
  * SQLite's busy result-code family — base `SQLITE_BUSY` plus the extended