mosquitto_loop_stop() hangs on windows

[ralight]

migrated from Bugzilla #452918
status NEW severity normal in component Mosquitto client library for ---
Reported in version unspecified on platform PC
Assigned to: Roger Light

On 2014-11-23 16:29:33 -0500, Roger Light wrote:

(imported from launchpad)

1. On many unix-like systems, system call such as read, write is a thread cancellation point specified by posix. But on windows, pthread_win32 implementation, socket family system call is not thread cancellation point, that means if i enable the thread support in mosquitto, there wouldn't be any thread cacellation point in mosquitto_loop_forever. So if calling mosquitto_loop_stop(mosq, true), then the pthread_join() which is called after pthread_cancel() will be blocked as the mosquitto_loop_forever() has not any cancellation point.

a patch to lib/mosquitto.c below to setup a cancellation point manually:

@@ -868,6 +868,9 @@ int mosquitto_loop_forever(struct mosquitto *mosq, int timeout, int max_packets)

while(run){
do{
+#ifdef WIN32

• pthread_testcancel();
  +#endif
  rc = mosquitto_loop(mosq, timeout, max_packets);
  if (reconnects !=0 && rc == MOSQ_ERR_SUCCESS){
  reconnects = 0;

1. Sleep() call on windows is not a cancellation point, too. And it is used in mosquitto_loop_forever() to do backoff in reconnecting phase, it will also prevent mosquitto_loop_stop() from returning immediately. pthread_cond_timewait() may be a better alternative.

[SigmundVik]

Excellent, thank you for the fix 👍

[ralight]

No problem. I just didn't have the time to be happy with your other patch, but I've not forgotten it.