SPL Meter 0.4.3: a failed capture-tool build must not park the listener forever

Same deadlock the Timers module hit live on AB (fixed there in 1.5.1):
start() skipped scheduling its retry while a build was in flight, and
ensureCaptureBinary's onReady only fires on build SUCCESS. A failed
build therefore left the listener waiting forever on a callback that
never comes — "Building the capture tool…" indefinitely — and a binary
scp'd in from outside (the required path on AB, whose CLT cannot
compile) was never picked up without a module remount.

onReady stays as the fast path; the 30-second retry is now armed in
every not-ready state, so a failed build keeps re-checking and an
externally supplied binary is found within half a minute.

Reproduced pre-fix (build fails at 0s, binary copied in at 3s, still
untouched at 38s) and verified post-fix (retry at 30s spawns it).
spl-capture --selftest unchanged against the IEC 61672 table.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

modules/spl-meter/module.json

@@ -1,7 +1,7 @@
 {
   "id": "spl-meter",
   "name": "SPL Meter",
-  "version": "0.4.1",
+  "version": "0.4.3",
   "proddash": ">=1.5.0",
   "description": "Calibrated sound level, Leq and RTA from an audio input on this machine",
   "permissions": [

modules/spl-meter/server.js

@@ -20,7 +20,7 @@
  */
 
 const path = require('path');
-const { createSplListener } = require('./spl-listener');
+const { createSplListener, ensureCaptureBinary } = require('./spl-listener');
 
 /** Reference level of a sound level calibrator. Effectively always 94 dB. */
 const CALIBRATOR_DB = 94;
@@ -114,11 +114,19 @@ module.exports = {
       error: '',
       streams: new Set(),
       silentSince: null,
+      toolMissing: '',
       listener: null,
       lastSent: '',
       pinger: null,
     };
 
+    // Before anything else: the device picker shells out to this tool, so a
+    // machine that installed the module from the catalog (source only — the
+    // binary is gitignored) needs the build kicked off here. Waiting for a
+    // listener would never happen: no tool means no device list, and no
+    // device means no listener.
+    state.toolMissing = ensureCaptureBinary({ moduleDir, log });
+
     if (!config.enabled) {
       state.error = 'Sound level measurement is off — turn it on in ProdDash settings';
       return { stop() { teardown(); }, health: () => ({ status: 'error', message: state.error }) };
@@ -177,6 +185,9 @@ module.exports = {
     return {
       stop() { teardown(); },
       health() {
+        if (state.toolMissing && !fsExists(path.join(moduleDir, 'spl-capture'))) {
+          return { status: 'error', message: state.toolMissing };
+        }
         if (state.error) return { status: 'error', message: state.error };
         if (!state.frame) return { status: 'connecting', message: 'waiting for the first reading' };
         const quiet = silentMs();
@@ -300,6 +311,13 @@ module.exports = {
       /** Device list for the admin page. Runs the capture tool's --list,
           which is the only thing that knows what CoreAudio can see. */
       'GET /devices': async (req, res) => {
+        const missing = ensureCaptureBinary({ moduleDir: __dirname, log });
+        if (missing) {
+          if (state) state.toolMissing = missing;
+          log(`device list unavailable — ${missing}`);
+          json(res, 200, { options: [] });
+          return;
+        }
         try {
           const devices = await listDevices();
           json(res, 200, {
@@ -438,6 +456,10 @@ function listDevices() {
   });
 }
 
+function fsExists(p) {
+  try { return require('fs').existsSync(p); } catch { return false; }
+}
+
 function json(res, code, body) {
   const text = JSON.stringify(body);
   res.writeHead(code, {

modules/spl-meter/spl-listener.js

@@ -17,7 +17,8 @@
  * module's ltc-capture. It normally arrives already built — compiled beside
  * the app, or copied from any Mac of the same architecture. When it is
  * missing this will build it, but in the background and never on the event
- * loop: see ensureBinary for what that cost the first time it was wrong.
+ * loop, and from module scope so the device picker can trigger it too — see
+ * ensureCaptureBinary for what both of those cost the first time round.
  *
  * The same macOS gotcha applies as for LTC, and is worth repeating because
  * it costs an afternoon every time: a ProdDash launched over SSH or from a
@@ -30,6 +31,74 @@ const path = require('path');
 const fs = require('fs');
 const { spawn } = require('child_process');
 
+const BUILD_HELP = 'Sound level capture tool missing — build it on any Apple Silicon Mac with '
+  + '`swiftc -O -o modules/spl-meter/spl-capture modules/spl-meter/spl-capture.swift` '
+  + 'and copy it next to spl-capture.swift';
+const BUILDING = 'Building the capture tool — this takes a minute or two';
+
+/**
+ * Build the capture tool if it is missing, WITHOUT blocking, and shared by
+ * everything that needs it.
+ *
+ * This lives at module scope rather than inside a listener because of a trap
+ * the first version walked into: the admin page's device picker runs the tool
+ * with --list, but the build only ran when a listener started, and a listener
+ * only starts once a device has been picked. On a machine that got the module
+ * from the catalog — source only, the binary being gitignored — the picker
+ * was empty, and nothing would ever fill it.
+ *
+ * It also must never block. It was a spawnSync once, copied from the LTC
+ * listener where the source is a third the size and builds in seconds. Here
+ * it took nearly two minutes on the production Mac and held the Node event
+ * loop for all of it, freezing every other module, then did it again on the
+ * next retry.
+ *
+ * The binary should normally already be there: built beside the app, or
+ * copied from any Mac of the same architecture. This is the fallback.
+ *
+ * @returns {string} '' when the tool is ready, otherwise what to say.
+ */
+let building = false;
+let buildFailed = false;
+
+function ensureCaptureBinary({ moduleDir, log, onReady }) {
+  const bin = path.join(moduleDir, 'spl-capture');
+  if (fs.existsSync(bin)) return '';
+  const src = bin + '.swift';
+  if (!fs.existsSync(src)) return 'spl-capture.swift is missing from the module folder';
+  if (building) return BUILDING;
+  if (buildFailed) return BUILD_HELP;
+
+  building = true;
+  log('spl-capture binary missing — building it in the background');
+  const build = spawn('swiftc', ['-O', '-o', bin, src], { stdio: ['ignore', 'ignore', 'pipe'] });
+  let buildErr = '';
+  build.stderr.setEncoding('utf8');
+  build.stderr.on('data', (t) => { buildErr = (buildErr + t).slice(-800); });
+  const giveUp = setTimeout(() => {
+    try { build.kill('SIGKILL'); } catch { /* already gone */ }
+  }, 300000);
+  build.on('error', (err) => {
+    clearTimeout(giveUp);
+    building = false; buildFailed = true;
+    log(`swiftc could not be run: ${err.message}`);
+  });
+  build.on('exit', (code) => {
+    clearTimeout(giveUp);
+    building = false;
+    if (code === 0 && fs.existsSync(bin)) {
+      log('spl-capture built');
+      buildFailed = false;
+      if (typeof onReady === 'function') onReady();
+      return;
+    }
+    buildFailed = true;
+    log(`spl-capture build failed (exit ${code})`
+      + (buildErr ? `: ${buildErr.trim().split('\n').pop()}` : ''));
+  });
+  return BUILDING;
+}
+
 const RESTART_MS = 5000;
 /** An immediate exit is config-shaped (bad device, channel out of range) —
     retrying every 5 s would only fill the log with the same sentence. */
@@ -53,64 +122,13 @@ function createSplListener(opts) {
   let stallTimer = null;
   let startedAt = 0;
   let meta = null;
-  let building = false;
-  let buildFailed = false;
   let lastFrameWall = 0;
 
-  const BUILD_HELP = 'Sound level capture tool missing — build it on any Apple Silicon Mac with '
-    + '`swiftc -O -o modules/spl-meter/spl-capture modules/spl-meter/spl-capture.swift` '
-    + 'and copy it next to spl-capture.swift';
-
-  /**
-   * Build the capture tool if it is missing, WITHOUT blocking.
-   *
-   * This used to be a spawnSync, copied from the LTC listener where the
-   * source is a third the size and builds in a few seconds. Here it took
-   * nearly two minutes on the production Mac and froze the entire ProdDash
-   * server for every second of it — one module's optional helper took the
-   * whole dashboard down, then did it again on every retry. Nothing that
-   * runs while a service might be on is allowed to hold the event loop.
-   *
-   * The build is a fallback for a module installed from the catalog on a
-   * machine with a Swift toolchain. The normal path is a binary built
-   * alongside the app, or copied from any same-architecture Mac.
-   *
-   * @returns {string} '' when the binary is ready, otherwise what to say.
-   */
+  /** '' when the tool is ready, otherwise what to say about it. */
   function ensureBinary() {
-    if (fs.existsSync(bin)) return '';
-    const src = bin + '.swift';
-    if (!fs.existsSync(src)) return 'spl-capture.swift is missing from the module folder';
-    if (building) return 'Building the capture tool — this takes a minute or two';
-    if (buildFailed) return BUILD_HELP;
-
-    building = true;
-    log('spl-capture binary missing — building it in the background');
-    const build = spawn('swiftc', ['-O', '-o', bin, src], { stdio: ['ignore', 'ignore', 'pipe'] });
-    let buildErr = '';
-    build.stderr.setEncoding('utf8');
-    build.stderr.on('data', (t) => { buildErr = (buildErr + t).slice(-800); });
-    const giveUp = setTimeout(() => {
-      try { build.kill('SIGKILL'); } catch { /* already gone */ }
-    }, 300000);
-    build.on('error', (err) => {
-      clearTimeout(giveUp);
-      building = false; buildFailed = true;
-      log(`swiftc could not be run: ${err.message}`);
-    });
-    build.on('exit', (code) => {
-      clearTimeout(giveUp);
-      building = false;
-      if (code === 0 && fs.existsSync(bin)) {
-        log('spl-capture built');
-        buildFailed = false;
-        if (!stopped) { clearTimeout(restartTimer); restartTimer = null; start(); }
-        return;
-      }
-      buildFailed = true;
-      log(`spl-capture build failed (exit ${code})` + (buildErr ? `: ${buildErr.trim().split('\n').pop()}` : ''));
-    });
-    return 'Building the capture tool — this takes a minute or two';
+    return ensureCaptureBinary({ moduleDir, log, onReady() {
+      if (!stopped) { clearTimeout(restartTimer); restartTimer = null; start(); }
+    } });
   }
 
   function scheduleRestart(ms) {
@@ -135,9 +153,13 @@ function createSplListener(opts) {
     const missing = ensureBinary();
     if (missing) {
       onUpdate({ error: missing });
-      // A build in flight calls start() itself when it lands; anything else
-      // is config-shaped and should back off hard rather than spin.
-      if (!building) scheduleRestart(BAD_CONFIG_RETRY_MS);
+      // onReady is only the FAST path (a successful build restarts us at
+      // once). The slow retry must stay armed regardless: a failed build
+      // never calls back, and a binary can also arrive from outside — scp'd
+      // in on a machine whose toolchain can't build (AB's broken CLT).
+      // Waiting on onReady alone deadlocked exactly that way in the Timers
+      // module: "building…" forever, with a working binary on disk.
+      scheduleRestart(BAD_CONFIG_RETRY_MS);
       return;
     }
 
@@ -236,4 +258,4 @@ function createSplListener(opts) {
   };
 }
 
-module.exports = { createSplListener };
+module.exports = { createSplListener, ensureCaptureBinary };