Pit toggle: ride out the stale-helper window after launch instead of failing (#103)

First click on bonita after tron upgrade: 'The pit resolver couldn't start', and the helper log showed the 3.3.0 helper listening with no pit: line at all. The request had reached the OLD 3.2.8 helper, which the launcher replaces in the background a moment after launch; it answered 404 {error:not-found} and the sidebar showed the generic message.

The background now retries /pit/start every 2s for up to 5 attempts while the answer is a 404 or nobody is listening, then reports helper-stale; the sidebar names that case (try again, or quit and relaunch) and shows the error code on the generic fallback. The helper logs unknown routes so a stale-version hit leaves a trace next time. Helper version 3.3.1.

Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com>

Author: ralyodio

apps/desktop/extensions/ai-sidebar/background.js

@@ -338,6 +338,34 @@ async function disablePit() {
   await setPitBadge(false);
 }
 
+// The launcher replaces an out-of-date helper in the background right after
+// launch (kill, settle, exec). A click in that window reaches the OLD helper,
+// which has no /pit/* routes and answers 404 {"error":"not-found"}, or reaches
+// nobody at all. Both are transient, so try for a few seconds before giving up,
+// and say "stale helper" rather than "couldn't start" when it never catches up.
+const PIT_START_ATTEMPTS = 5;
+const PIT_START_RETRY_MS = 2000;
+
+function pitHelperIsStale(res) {
+  return !res || res.error === 'not-found' || (!res.started && typeof res.port !== 'number');
+}
+
+async function startPitViaHelper() {
+  let last = null;
+  for (let i = 0; i < PIT_START_ATTEMPTS; i++) {
+    if (i) await new Promise((r) => setTimeout(r, PIT_START_RETRY_MS));
+    try {
+      last = await helperJson('/pit/start', 'POST');
+    } catch (_) {
+      last = { error: 'unreachable' };
+      continue;
+    }
+    if (!pitHelperIsStale(last)) return last; // a pit-capable helper answered, ok or not
+  }
+  if (last && last.error === 'unreachable') return last;
+  return { started: false, error: 'helper-stale' };
+}
+
 async function stopPitViaHelper() {
   try {
     const ctrl = new AbortController();
@@ -356,15 +384,9 @@ chrome.runtime.onMessage.addListener((msg, _sender, sendResponse) => {
           sendResponse({ enabled: false, error: 'tor-on' });
           return;
         }
-        let started;
-        try {
-          started = await helperJson('/pit/start', 'POST');
-        } catch (_) {
-          sendResponse({ enabled: false, error: 'unreachable' });
-          return;
-        }
-        if (!started || !started.started) {
-          sendResponse({ enabled: false, error: (started && started.error) || 'pit-failed' });
+        const started = await startPitViaHelper();
+        if (!started.started) {
+          sendResponse({ enabled: false, error: started.error || 'pit-failed' });
           return;
         }
         await enablePit();

apps/desktop/extensions/ai-sidebar/sidepanel.js

@@ -337,8 +337,10 @@ async function togglePit() {
         showNetStatus('warn', 'Couldn’t reach the TronBrowser helper. Restart TronBrowser and try again, or run <code>tron upgrade</code>.');
       } else if (err === 'pit-port-busy') {
         showNetStatus('warn', `Port ${PIT_SOCKS_PORT} on this machine is taken by another program, so the pit resolver couldn’t start.`);
+      } else if (err === 'helper-stale') {
+        showNetStatus('warn', 'TronBrowser’s helper is still the older version, which has no pit resolver. It is normally replaced a few seconds after launch — try 🤘 Pit again; if it keeps failing, quit TronBrowser completely and relaunch it.');
       } else {
-        showNetStatus('warn', 'The pit resolver couldn’t start. See <code>~/.tronbrowser/tor-helper.log</code> for the reason.');
+        showNetStatus('warn', `The pit resolver couldn’t start (${safeHost(err || 'pit-failed')}). See <code>~/.tronbrowser/tor-helper.log</code> for the reason.`);
       }
     }
   } catch (e) {

apps/desktop/launcher/tron-tor-helper

@@ -49,7 +49,7 @@ BUNDLED_DIR = os.environ.get("TRON_TOR_BIN_DIR", "")
 PIDFILE = os.environ.get("TRON_TOR_PIDFILE", "")
 # Bumped whenever the helper protocol/behaviour changes; the launcher kills a
 # stale helper so the current version always runs.
-HELPER_VERSION = "3.3.0"
+HELPER_VERSION = "3.3.1"
 _lock = threading.Lock()
 _proc = None              # the running tor subprocess (or None)
 _ready = False            # True once tor reported Bootstrapped 100%
@@ -559,6 +559,7 @@ class Handler(BaseHTTPRequestHandler):
         elif path == "/pit/status":
             self._send(200, pit_status())
         else:
+            log("no such route: %s (helper v%s)" % (path, HELPER_VERSION))
             self._send(404, {"error": "not-found"})
 
     def do_GET(self):

apps/desktop/launcher/tronbrowser

@@ -175,7 +175,7 @@ if [ "$TOR" != "1" ]; then
     # running helper isn't this version — otherwise leave a healthy current
     # helper alone (don't drop an active Tor session). All backgrounded so the
     # kill+settle never holds up the browser launch.
-    HELPER_VERSION=3.3.0
+    HELPER_VERSION=3.3.1
     (
       _pf="$DATA/tor-helper.pid"
       _rv="$(curl -fsS --max-time 1 http://127.0.0.1:9061/status 2>/dev/null | sed -n 's/.*"version"[^"]*"\([^"]*\)".*/\1/p')"

docs/moshpit-pit-toggle.md

@@ -75,7 +75,11 @@ would leak every lookup outside Tor, so:
 - If the DoH resolver did not answer the probe (offline, slow), the pit stays
   **on** and says so; names resolve as soon as it is reachable.
 - Failures the sidebar explains: the helper is not running (`tron upgrade`,
-  restart), port 9081 is taken, Tor is on.
+  restart), port 9081 is taken, Tor is on, or the helper is still an older
+  version. That last one is a launch race: the launcher replaces an out-of-date
+  helper in the background a moment after the browser starts, and a click in
+  that window reaches the old helper's 404. The background retries `/pit/start`
+  for about ten seconds before reporting it.
 
 ## Files