fix(shutdown): hold the signal listener until the exit decision is made

Closing the last live instance inside the SIGTERM/SIGINT handler calls
close() -> deregisterShutdownHooksIfIdle(), which removes Brainy's own
signal listener from process synchronously, before that same handler
invocation has reached the point where it decides whether to exit.
That opens a window with no registered listener for the signal at
all: a second/concurrent delivery of the same signal during that
window falls through to Node's default disposition and kills the
process outright, after the clean shutdown already finished, so the
process reports a signal kill instead of the 0 clause (a) and (b) of
shutdown-single-owner.test.ts pin — intermittent under load, which is
why it only ever showed up on the box.

Add a static flag that stays true for the whole closeOnShutdown() run
and makes deregisterShutdownHooksIfIdle() defer rather than remove the
listener while that run is still deciding; closeOnShutdown()'s own
finally re-runs the deregistration check once it is actually done, so
the listener never leaks past its use.

scripts/wall-entry.mjs

@@ -337,6 +337,36 @@ function git(args, cwd) {
   }
 }
 
+/**
+ * Resolve the git identity for the wall commit from the repository the rail
+ * is actually running in — the developer's own checkout (`process.cwd()`;
+ * `release.sh` invokes this script from the repo root with no `cd`), via
+ * git's normal config precedence (repo-local, then global, then system).
+ * Never guessed and never left to git's own "who are you?" prompt: a host
+ * with no configured identity anywhere (a bare CI box, say) must refuse
+ * loudly rather than have git manufacture a placeholder identity or hang.
+ * @returns {{name: string, email: string}}
+ */
+function resolveWallCommitIdentity() {
+  const repo = process.cwd()
+  let name = ''
+  let email = ''
+  try {
+    name = git(['config', 'user.name'], repo)
+  } catch {
+    name = ''
+  }
+  try {
+    email = git(['config', 'user.email'], repo)
+  } catch {
+    email = ''
+  }
+  if (!name || !email) {
+    fail('no git identity for the wall commit — set user.name/user.email')
+  }
+  return { name, email }
+}
+
 /**
  * Ensure a clean, up-to-date local clone of the releases repo at
  * `cacheDir`, checked out on `main` — cloning fresh if `cacheDir` has no
@@ -423,9 +453,14 @@ function publishEntry(entry, product, remote, cacheDir) {
     return
   }
 
+  const identity = resolveWallCommitIdentity()
+
   try {
     git(['add', `${product}.json`], cacheDir)
-    git(['commit', '-m', `chore(wall): ${product} ${entry.version}`], cacheDir)
+    git(
+      ['-c', `user.name=${identity.name}`, '-c', `user.email=${identity.email}`, 'commit', '-m', `chore(wall): ${product} ${entry.version}`],
+      cacheDir,
+    )
   } catch (err) {
     fail(`cannot commit the wall entry in "${cacheDir}" — ${/** @type {Error} */ (err).message}\n  cure: inspect "${cacheDir}" by hand and re-run once its git state is clean`)
   }

src/brainy.ts

@@ -544,6 +544,23 @@ export class Brainy<T = any> implements BrainyInterface<T> {
    *  and narrate forever. Reset by {@link deregisterShutdownHooksIfIdle}. */
   private static beforeExitNarrated = false
 
+  /** True for the entire duration of ONE `closeOnShutdown()` run (the
+   *  signal-path handler in {@link registerShutdownHooks}) — from before it
+   *  starts closing instances until after it has decided whether to exit.
+   *  THE RACE THIS CLOSES: closing the LAST live instance calls
+   *  `close()` → `deregisterShutdownHooksIfIdle()` synchronously, which
+   *  removes `Brainy.sigtermListener` from `process` — while `closeOnShutdown`
+   *  (that very listener's OWN still-running invocation) hasn't yet reached
+   *  `exitIfSoleShutdownOwner()`'s `process.exit(0)`. In that window Node has
+   *  NO registered SIGTERM listener, so a second/concurrent delivery of the
+   *  same signal (a raced re-send, common on a loaded host) falls through to
+   *  Node's default disposition and kills the process outright — the
+   *  clean-shutdown work already finished, but the process never reports the
+   *  0 it earned. `deregisterShutdownHooksIfIdle()` checks this flag and
+   *  defers; `closeOnShutdown()`'s `finally` re-runs the deregistration check
+   *  once it is done, so the listener never actually leaks past its use. */
+  private static shutdownSignalHandlerActive = false
+
   /** Poll cadence (ms) for the migration LOCK when a provider exposes no
    *  event-driven `whenMigrationComplete()` signal. See {@link awaitMigrationLock}. */
   private static readonly MIGRATION_POLL_INTERVAL_MS = 250
@@ -2196,13 +2213,29 @@ export class Brainy<T = any> implements BrainyInterface<T> {
      */
     const closeOnShutdown = async () => {
       console.log('Shutdown signal received - flushing pending data...')
-      // DEFER ONE MACROTASK. A host application registers its own listener on
-      // the same signal, and Node runs listeners in registration order — ours
-      // is usually first, because the brain was opened before the host wired
-      // its shutdown. Yielding once lets every other listener for this signal
-      // run its synchronous prologue, so a host that calls close() gets to be
-      // the owner. It is only a courtesy, never the safety: close()'s own
-      // single-flight gate is what makes a lost race harmless.
+      // HOLD THE LISTENER FOR THE WHOLE RUN. Closing the LAST live instance
+      // below calls close() → deregisterShutdownHooksIfIdle(), which removes
+      // Brainy's own SIGTERM/SIGINT listeners from `process` — synchronously,
+      // before THIS invocation has reached exitIfSoleShutdownOwner()'s
+      // process.exit(0). Left alone, that opens a window with no registered
+      // listener for the signal at all, so a second/concurrent delivery of
+      // the same signal (a raced re-send — not rare on a loaded host) falls
+      // through to Node's default disposition and kills the process outright
+      // AFTER the clean-shutdown work already finished, reporting a signal
+      // kill instead of the 0 the shutdown earned. Setting this flag makes
+      // deregisterShutdownHooksIfIdle() defer; the `finally` below re-checks
+      // it once this run is fully done — closeOnShutdown, not a nested
+      // close(), owns exactly when the listener actually comes off.
+      Brainy.shutdownSignalHandlerActive = true
+      try {
+        // DEFER ONE MACROTASK. A host application registers its own listener
+        // on the same signal, and Node runs listeners in registration order —
+        // ours is usually first, because the brain was opened before the
+        // host wired its shutdown. Yielding once lets every other listener
+        // for this signal run its synchronous prologue, so a host that calls
+        // close() gets to be the owner. It is only a courtesy, never the
+        // safety: close()'s own single-flight gate is what makes a lost race
+        // harmless.
         await new Promise<void>((resolve) => setImmediate(resolve))
 
         let closedCount = 0
@@ -2242,6 +2275,13 @@ export class Brainy<T = any> implements BrainyInterface<T> {
               `their writer locks were released, but their next open will run crash recovery.`
           )
         }
+      } finally {
+        // Release the hold and run the deferred check ourselves — the last
+        // close() above may have found the flag set and skipped its own
+        // deregistration, so nobody else will do this if we don't.
+        Brainy.shutdownSignalHandlerActive = false
+        Brainy.deregisterShutdownHooksIfIdle()
+      }
     }
 
     /**
@@ -2404,9 +2444,17 @@ export class Brainy<T = any> implements BrainyInterface<T> {
    * script that closed every brain exits on its own — a library must never
    * keep its host process alive. Re-initializing later re-registers them
    * (the `shutdownHooksRegisteredGlobally` flag resets here).
+   *
+   * Deferred (not skipped — {@link closeOnShutdown}'s `finally` always
+   * re-checks) while a signal-path shutdown is actively running: that
+   * handler's OWN still-in-flight invocation is `Brainy.sigtermListener`, and
+   * removing it out from under itself — which closing the LAST instance here
+   * would otherwise do, synchronously, mid-run — would leave `process` with
+   * no listener for the signal for the remainder of that run. See
+   * {@link shutdownSignalHandlerActive}'s doc for the exact race this closes.
    */
   private static deregisterShutdownHooksIfIdle(): void {
-    if (Brainy.instances.length > 0 || !Brainy.shutdownHooksRegisteredGlobally) {
+    if (Brainy.instances.length > 0 || !Brainy.shutdownHooksRegisteredGlobally || Brainy.shutdownSignalHandlerActive) {
       return
     }
     if (Brainy.sigtermListener) process.off('SIGTERM', Brainy.sigtermListener)

tests/unit/release/wall-entry.test.ts

@@ -104,6 +104,14 @@ let cacheDir: string
 
 beforeEach(() => {
   dir = mkdtempSync(join(tmpdir(), 'wall-entry-test-'))
+  // wall-entry.mjs is run with this dir as its cwd, standing in for the real
+  // developer checkout it reads its commit identity from (process.cwd()) —
+  // give it a repo-local identity the same way seedRemote gives one to the
+  // seed clone, so the suite is deterministic on a host with no global git
+  // config (a bare CI box) as much as one with a developer's own.
+  execFileSync('git', ['init', '-q', dir])
+  git(['config', 'user.name', 'Wall Entry Test'], dir)
+  git(['config', 'user.email', 'wall-entry-test@example.com'], dir)
   remoteDir = initBareRemote()
   cacheDir = join(mkdtempSync(join(tmpdir(), 'wall-cache-')), 'soulcraft-releases')
 })