Update check: say what the forge answered, not zlib's "incorrect header check"

A Gitea that requires a sign-in does not refuse the branch archive: it
redirects the download to its login page, which arrives as a 200 of HTML,
and the admin page then reported zlib failing to unpack it — "incorrect
header check". Both booths showed exactly that on 29 Sept once the sign-in
requirement was switched on.

The updater now names the case: "… requires a sign-in to read
michael/ProdDash@main — turn off 'require sign-in to view' in Gitea, or
make the repo readable without one". Any other answer that is not a gzip
archive says what came back instead — a web page, or the content type —
and where to look.

tools/update-check.test.js drives the check against a mock forge: the login
redirect, a bare HTML answer, an HTTP error, and a real archive that must
still parse and be offered.

Shell 1.9.3.

Co-Authored-By: Claude Fable 5.1 <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;
+    return ensureCaptureBinary({ moduleDir, log, onReady() {
       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';
+    } });
   }
 
   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 };

package.json

@@ -1,6 +1,6 @@
 {
   "name": "proddash",
-  "version": "1.9.2",
+  "version": "1.9.3",
   "private": true,
   "description": "Modular production dashboard for Waters Church \u2014 zero-dependency Node server + vanilla client",
   "license": "UNLICENSED",

server.js

@@ -910,9 +910,21 @@ async function fetchArchive() {
     signal: AbortSignal.timeout(30000),
     headers: { 'User-Agent': `ProdDash/${SHELL_VERSION}` },
   });
-  if (!res.ok) throw new Error(`HTTP ${res.status} fetching ${REPO.owner}/${REPO.repo}@${REPO.branch}`);
+  const where = `${REPO.owner}/${REPO.repo}@${REPO.branch}`;
+  // A forge that wants a sign-in does not refuse: it redirects the download
+  // to its login page, which arrives as a 200 of HTML. Say that, not
+  // zlib's "incorrect header check".
+  if (res.redirected && /\/user\/login\b|\/login\b/.test(new URL(res.url).pathname)) {
+    throw new Error(`${FORGE.base} requires a sign-in to read ${where} — turn off "require sign-in to view" in Gitea, or make the repo readable without one`);
+  }
+  if (!res.ok) throw new Error(`HTTP ${res.status} fetching ${where}`);
   const gz = Buffer.from(await res.arrayBuffer());
   if (gz.length > MAX_ARCHIVE_BYTES) throw new Error('archive is unexpectedly large');
+  if (gz.length < 2 || gz[0] !== 0x1f || gz[1] !== 0x8b) {
+    const type = String(res.headers.get('content-type') || 'unknown type').split(';')[0].trim();
+    const looksHtml = /^\s*<(!doctype|html)/i.test(gz.subarray(0, 64).toString('utf8'));
+    throw new Error(`${FORGE.base} answered with ${looksHtml ? 'a web page' : type} instead of the ${where} archive — check the forge address and that the repo is readable without a sign-in`);
+  }
   const { files, comment } = parseTar(zlib.gunzipSync(gz));
   const stripped = new Map(); // drop the archive's single top-level folder
   for (const [p, data] of files) {

tools/update-check.test.js

@@ -0,0 +1,176 @@
+'use strict';
+
+/**
+ * The update check against a forge that misbehaves: the admin page must say
+ * what went wrong in words, never zlib's "incorrect header check".
+ *
+ *   node tools/update-check.test.js
+ *
+ * Plain Node, no dependencies. Starts a mock forge whose answer to the
+ * branch-archive download can be switched at runtime, starts server.js on a
+ * spare port pointed at it, and reads the shell status the admin page shows
+ * after each kind of answer:
+ *
+ *   - a Gitea that requires a sign-in redirects the download to its login
+ *     page (a 200 of HTML) — the case that hit both booths on 2026-09-29;
+ *   - a captive portal or a wrong address answers HTML outright;
+ *   - a plain HTTP error;
+ *   - and a real archive, which must still parse and offer the update.
+ *
+ * Exits non-zero on the first failure and always stops what it started.
+ */
+
+const assert = require('node:assert/strict');
+const { spawn, execFileSync } = require('node:child_process');
+const fs = require('node:fs');
+const http = require('node:http');
+const net = require('node:net');
+const os = require('node:os');
+const path = require('node:path');
+
+const ROOT = path.join(__dirname, '..');
+
+function freePort() {
+  return new Promise((resolve, reject) => {
+    const probe = net.createServer();
+    probe.listen(0, '127.0.0.1', () => {
+      const { port } = probe.address();
+      probe.close(() => resolve(port));
+    });
+    probe.on('error', reject);
+  });
+}
+
+async function waitFor(url, ms = 15000) {
+  const deadline = Date.now() + ms;
+  let last = '';
+  while (Date.now() < deadline) {
+    try {
+      const r = await fetch(url);
+      if (r.ok) return;
+      last = 'HTTP ' + r.status;
+    } catch (err) {
+      last = err.message;
+    }
+    await new Promise((r) => setTimeout(r, 200));
+  }
+  throw new Error('server did not come up: ' + last);
+}
+
+/** A branch archive the way a forge would build one: one top-level folder. */
+function buildArchive(dir) {
+  const top = path.join(dir, 'ProdDash-main');
+  fs.mkdirSync(path.join(top, 'modules', 'demo'), { recursive: true });
+  fs.writeFileSync(path.join(top, 'package.json'), JSON.stringify({ name: 'proddash', version: '9.9.9' }));
+  fs.writeFileSync(path.join(top, 'modules', 'demo', 'module.json'), JSON.stringify({ id: 'demo', name: 'Demo', version: '1.0.0', client: 'client.js' }));
+  fs.writeFileSync(path.join(top, 'modules', 'demo', 'client.js'), 'export default function () {}\n');
+  const out = path.join(dir, 'main.tar.gz');
+  execFileSync('tar', ['-czf', out, '-C', dir, 'ProdDash-main']);
+  return fs.readFileSync(out);
+}
+
+(async () => {
+  const work = fs.mkdtempSync(path.join(os.tmpdir(), 'proddash-update-test-'));
+  const dataDir = path.join(work, 'data');
+  fs.mkdirSync(dataDir);
+  const archive = buildArchive(path.join(work, 'forge'));
+
+  // ── the mock forge ──
+  let mode = 'ok';
+  const LOGIN_PAGE = '<!DOCTYPE html>\n<html><head><title>Sign in</title></head><body>Sign in to continue</body></html>';
+  const forge = http.createServer((req, res) => {
+    const url = new URL(req.url, 'http://forge.invalid');
+    if (url.pathname === '/user/login') {
+      res.writeHead(200, { 'Content-Type': 'text/html; charset=utf-8' });
+      return res.end(LOGIN_PAGE);
+    }
+    if (url.pathname !== '/michael/ProdDash/archive/main.tar.gz') {
+      res.writeHead(404, { 'Content-Type': 'text/plain' });
+      return res.end('not here');
+    }
+    switch (mode) {
+      case 'login-redirect':
+        res.writeHead(303, { Location: '/user/login?redirect_to=' + encodeURIComponent(url.pathname) });
+        return res.end();
+      case 'html':
+        res.writeHead(200, { 'Content-Type': 'text/html; charset=utf-8' });
+        return res.end('<html><body>You must log in to this network first</body></html>');
+      case 'notfound':
+        res.writeHead(404, { 'Content-Type': 'text/plain' });
+        return res.end('no such branch');
+      default:
+        res.writeHead(200, { 'Content-Type': 'application/gzip' });
+        return res.end(archive);
+    }
+  });
+  const forgePort = await freePort();
+  await new Promise((r) => forge.listen(forgePort, '127.0.0.1', r));
+  const forgeBase = `http://127.0.0.1:${forgePort}`;
+
+  // ── ProdDash, pointed at it ──
+  const port = await freePort();
+  const child = spawn(process.execPath, ['server.js'], {
+    cwd: ROOT,
+    env: {
+      ...process.env,
+      PRODDASH_DATA_DIR: dataDir,
+      PORT: String(port),
+      PRODDASH_FORGE: forgeBase,
+      PRODDASH_REPO: 'michael/ProdDash',
+      PRODDASH_NO_REMOTE_CHECK: '1',
+      PRODDASH_LAUNCHER: '1',
+    },
+    stdio: ['ignore', 'pipe', 'pipe'],
+  });
+  let log = '';
+  child.stdout.on('data', (d) => { log += d; });
+  child.stderr.on('data', (d) => { log += d; });
+  const base = `http://127.0.0.1:${port}`;
+
+  /** What the admin page shows after a forced check. Admin is open: a fresh data dir has no PIN. */
+  const check = async (m) => {
+    mode = m;
+    const r = await fetch(base + '/api/admin/catalog?refresh=1');
+    assert.equal(r.status, 200, `catalog: HTTP ${r.status}`);
+    const body = await r.json();
+    return body.shell;
+  };
+
+  let failed = false;
+  try {
+    await waitFor(base + '/api/theme');
+
+    let shell = await check('login-redirect');
+    assert.match(shell.remote.error, /requires a sign-in/, 'a login redirect is named as such: ' + shell.remote.error);
+    assert.match(shell.remote.error, /require sign-in to view/, 'and the Gitea setting to look at is named');
+    assert.doesNotMatch(shell.remote.error, /header check/, 'never zlib\'s words');
+    assert.equal(shell.updateAvailable, false);
+
+    shell = await check('html');
+    assert.match(shell.remote.error, /answered with a web page instead of the michael\/ProdDash@main archive/, shell.remote.error);
+
+    shell = await check('notfound');
+    assert.match(shell.remote.error, /HTTP 404 fetching michael\/ProdDash@main/, shell.remote.error);
+
+    shell = await check('ok');
+    assert.equal(shell.remote.error, '', 'a real archive clears the error: ' + shell.remote.error);
+    assert.equal(shell.remote.version, '9.9.9', 'and is read');
+    assert.equal(shell.updateAvailable, true, 'and offered');
+
+    // Back to a broken forge: the last good read is not mistaken for a current one.
+    shell = await check('login-redirect');
+    assert.match(shell.remote.error, /requires a sign-in/);
+
+    console.log('update check: every check passed');
+  } catch (err) {
+    failed = true;
+    console.error('FAIL:', err.message);
+    console.error('--- last of the server log ---\n' + log.slice(-2000));
+  } finally {
+    child.kill('SIGTERM');
+    forge.close();
+    await new Promise((r) => setTimeout(r, 400));
+    fs.rmSync(work, { recursive: true, force: true });
+  }
+  process.exit(failed ? 1 : 0);
+})();