Fix macOS/Windows CI hang: correct second Select timeout, fix Cleanup lock scope, harden test thread

Agent-Logs-Url: https://github.com/zeromq/netmq/sessions/259a6416-41cd-429e-8110-4e1559b0130e

Co-authored-by: drewnoakes <[email protected]>

Author: Copilot

src/NetMQ.Tests/CleanupTests.cs

@@ -74,7 +74,10 @@ public void NoBlockCompletesInBoundedTime()
             // and an error list, causing Cleanup to hang forever without this guard.
             _ = new DealerSocket(">tcp://localhost:5557"); // intentionally not disposed
 
-            var thread = new Thread(() => NetMQConfig.Cleanup(block: false));
+            // Run cleanup on a background (daemon) thread so the process can still exit
+            // if the thread gets stuck. IsBackground = true prevents it from blocking
+            // process shutdown if a regression causes Cleanup to hang.
+            var thread = new Thread(() => NetMQConfig.Cleanup(block: false)) { IsBackground = true };
             thread.Start();
             Assert.True(thread.Join(TimeSpan.FromSeconds(10)),
                 "Cleanup(block: false) did not complete within 10 seconds");

src/NetMQ/Core/Utils/Poller.cs

@@ -290,7 +290,11 @@ private void Loop()
                         // together. To avoid this problem, we call the Select function separately for errorList.
                         // Please refer to this issue: https://github.com/dotnet/corefx/issues/39617
                         SocketUtility.Select(readList, null, null, timeout);
-                        SocketUtility.Select(null, null, errorList, timeout);
+                        // If the first select found readable sockets, use a non-blocking (0) timeout for the
+                        // error check so we don't block indefinitely on a stale list before processing InEvent.
+                        // Events queued by the first select (e.g. a ForceStop command) are not yet processed
+                        // because InEvent runs after both selects, so a second infinite-wait here would deadlock.
+                        SocketUtility.Select(null, null, errorList, readList.Count > 0 ? 0 : timeout);
                     }
                     else
                     {

src/NetMQ/NetMQConfig.cs

@@ -48,14 +48,17 @@ internal static Ctx Context
         /// <param name="block">Set to true when you want to make sure sockets send all pending messages</param>
         public static void Cleanup(bool block = true)
         {
+            Ctx? ctx;
             lock (s_sync)
             {
-                if (s_ctx != null)
-                {
-                    s_ctx.Terminate(block);
-                    s_ctx = null;
-                }
+                // Capture and clear the context reference while holding the lock, then
+                // call Terminate outside the lock so a long-running or stuck Terminate
+                // (e.g. on macOS with block:false) never prevents other threads from
+                // acquiring s_sync and observing that cleanup has already been initiated.
+                ctx = s_ctx;
+                s_ctx = null;
             }
+            ctx?.Terminate(block);
         }
 
         /// <summary>