Hello, list Please consider my application:
1. There is one process that always in event loop (master process). 2. By some event it forks and create "child" process, that call event_exit_loop() and after event_loop() exits it try to clean up everything associated with libevent. 3. For this purpose child call event_base_free(), but it have to remove all events that was added by parent before fork happened. Otherwise it will be hit by assert(TAILQ_EMPTY(&base->eventqueue)). 4. With poll or select methods there no problems with deleting events at this point. 5. But with epoll, kqueue or event ports methods the child and the parent shares *same* event device and set of file descriptors after fork(). So if child call event_del() on every event that was added before in parent, events are also deleted from parent's device (actually, the same device), which is absolutely not what I want. Thus, there is only one way to deal with it: child must not call event_base_free() and must live with opened event device. Attached diff provides a fix for this situation. I believe process must not check if base->eventqueue and other queues are empty. Calling evsel->dealloc() is enough to remove all events. -- Andrei Nigmatulin GPG PUB KEY 6449830D Now I lay me down to sleep(3) Pray the OS my core to keep If I die before I wake Pray the Disk my core to take
diff -burp libevent/event.c ./event.c --- libevent/event.c 2007-06-08 01:59:32.000000000 +0400 +++ ./event.c 2007-06-12 15:58:28.000000000 +0400 @@ -220,19 +220,14 @@ event_base_free(struct event_base *base) current_base = NULL; assert(base); + if (base->evsel->dealloc != NULL) base->evsel->dealloc(base, base->evbase); - for (i=0; i < base->nactivequeues; ++i) - assert(TAILQ_EMPTY(base->activequeues[i])); - - assert(RB_EMPTY(&base->timetree)); for (i = 0; i < base->nactivequeues; ++i) free(base->activequeues[i]); free(base->activequeues); - assert(TAILQ_EMPTY(&base->eventqueue)); - free(base); }
_______________________________________________ Libevent-users mailing list Libevent-users@monkey.org http://monkey.org/mailman/listinfo/libevent-users
