From d448dad390bea4daaab8424fd8b09dd823f23655 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E6=9D=A8=E5=98=89=E4=BC=9F?= <202383014@uibe.edu.cn>
Date: Tue, 22 Sep 2026 16:34:22 +0800
Subject: [PATCH] fix(setup): recover locks while writers wait

---
 extensions/shared/setup-config.ts            |   8 ++
 tests/extensions/shared/setup-config.test.ts | 106 ++++++++++++++++++-
 2 files changed, 113 insertions(+), 1 deletion(-)

diff --git a/extensions/shared/setup-config.ts b/extensions/shared/setup-config.ts
index 83ace16f..fac97947 100644
--- a/extensions/shared/setup-config.ts
+++ b/extensions/shared/setup-config.ts
@@ -219,6 +219,7 @@ export const DEFAULT_SETUP_CONFIG: MyPiSetupConfig = {
 export const SETUP_CONFIG_PATH = join(getAgentDir(), "my-pi-setup.json");
 const SETUP_CONFIG_LOCK_PATH = `${SETUP_CONFIG_PATH}.lock`;
 const SETUP_CONFIG_LOCK_TIMEOUT_MS = 5_000;
+const SETUP_CONFIG_LOCK_RECHECK_MS = 500;
 const SETUP_CONFIG_LOCK_VERSION = 1;
 const ESTIMATED_PROCESS_STARTED_AT = Math.max(
   1,
@@ -843,6 +844,12 @@ function waitForSetupConfigLock(deadline: number) {
   return new Promise<void>((resolve, reject) => {
     let settled = false;
     let watcher: ReturnType<typeof watch> | undefined;
+    // A dead owner does not change the lock file, so periodically retry the
+    // existing fail-closed recovery path before the shared deadline expires.
+    const recheckTimer = setTimeout(
+      () => finish(),
+      Math.min(remaining, SETUP_CONFIG_LOCK_RECHECK_MS),
+    );
     const timer = setTimeout(() => {
       // fs.watch may coalesce or drop events. Recheck the atomic lock path at
       // the deadline so a released lock cannot become a false timeout.
@@ -853,6 +860,7 @@ function waitForSetupConfigLock(deadline: number) {
       if (settled) return;
       settled = true;
       clearTimeout(timer);
+      clearTimeout(recheckTimer);
       watcher?.close();
       if (error) reject(error);
       else resolve();
diff --git a/tests/extensions/shared/setup-config.test.ts b/tests/extensions/shared/setup-config.test.ts
index 0984b962..b9c75157 100644
--- a/tests/extensions/shared/setup-config.test.ts
+++ b/tests/extensions/shared/setup-config.test.ts
@@ -5,8 +5,8 @@ import {
   existsSync,
   linkSync,
   mkdtempSync,
-  readFileSync,
   readdirSync,
+  readFileSync,
   rmSync,
   writeFileSync,
 } from "node:fs";
@@ -431,6 +431,110 @@ await updateSetupConfig((current) => {
   }
 });
 
+test("a waiting setup writer recovers after the lock owner dies", {
+  timeout: 15_000,
+}, async () => {
+  await saveSetupConfig(DEFAULT_SETUP_CONFIG);
+  const holderSource = `
+const { writeFileSync } = await import("node:fs");
+const { updateSetupConfig } = await import(process.env.SETUP_CONFIG_MODULE_URL);
+await updateSetupConfig((current) => {
+  writeFileSync(1, "holding\\n");
+  Atomics.wait(new Int32Array(new SharedArrayBuffer(4)), 0, 0);
+  return current;
+});
+`;
+  const waiterSource = `
+const fs = await import("node:fs");
+const { syncBuiltinESMExports } = await import("node:module");
+const originalWatch = fs.default.watch;
+fs.default.watch = (...args) => {
+  const watcher = originalWatch(...args);
+  process.stdout.write("watching\\n");
+  return watcher;
+};
+syncBuiltinESMExports();
+const { updateSetupConfig } = await import(process.env.SETUP_CONFIG_MODULE_URL);
+const started = Date.now();
+await updateSetupConfig((current) => current);
+process.stdout.write(JSON.stringify({ result: "success", ms: Date.now() - started }) + "\\n");
+`;
+  const start = (source: string) => {
+    const child = spawn(
+      process.execPath,
+      ["--experimental-strip-types", "--input-type=module", "--eval", source],
+      {
+        env: {
+          ...process.env,
+          PI_CODING_AGENT_DIR: agentDir,
+          SETUP_CONFIG_MODULE_URL: setupConfigModuleUrl,
+        },
+        stdio: ["ignore", "pipe", "pipe"],
+      },
+    );
+    child.stdout.setEncoding("utf8");
+    child.stderr.setEncoding("utf8");
+    let stdout = "";
+    let stderr = "";
+    child.stdout.on("data", (chunk: string) => {
+      stdout += chunk;
+    });
+    child.stderr.on("data", (chunk: string) => {
+      stderr += chunk;
+    });
+    const waitFor = (marker: string) =>
+      new Promise<void>((resolve, reject) => {
+        const check = () => {
+          if (!stdout.includes(marker)) return;
+          child.stdout.off("data", check);
+          resolve();
+        };
+        check();
+        child.stdout.on("data", check);
+        child.once("error", reject);
+        child.once("exit", (code) => {
+          if (!stdout.includes(marker))
+            reject(
+              new Error("child exited " + code + ": " + (stderr || stdout)),
+            );
+        });
+      });
+    return { child, waitFor, output: () => ({ stdout, stderr }) };
+  };
+
+  const holder = start(holderSource);
+  let waiter: ReturnType<typeof start> | undefined;
+  try {
+    await holder.waitFor("holding\n");
+    waiter = start(waiterSource);
+    await waiter.waitFor("watching\n");
+    const exited = once(holder.child, "exit");
+    holder.child.kill("SIGKILL");
+    await exited;
+
+    await once(waiter.child, "exit");
+    const { stdout, stderr } = waiter.output();
+    assert.equal(stderr, "");
+    const result = JSON.parse(stdout.split("\n").at(-2) ?? "");
+    assert.equal(result.result, "success");
+    assert.ok(
+      result.ms < 5_000,
+      "waiter used the full deadline: " + result.ms + "ms",
+    );
+  } finally {
+    for (const process of [holder, waiter]) {
+      if (!process) continue;
+      const { child } = process;
+      if (child.exitCode === null && child.signalCode === null) {
+        const exited = once(child, "exit");
+        child.kill("SIGKILL");
+        await exited;
+      }
+    }
+    removeLockArtifacts();
+  }
+});
+
 test("a reused PID does not impersonate the dead lock owner", async () => {
   removeLockArtifacts();
   await saveSetupConfig(DEFAULT_SETUP_CONFIG);