I'm implementing server which initializes listener like this (err handling omitted):

struct event_base* base = event_base_new();
struct sockaddr_in sin;
memset(&sin, 0, sizeof(sin));
sin.sin_family = AF_INET;
sin.sin_port = htons(8444);
sin.sin_addr.s_addr = htonl(INADDR_ANY);
struct evconnlistener* listener = evconnlistener_new_bind(base, HandleAccept, ctx,
                                       LEV_OPT_REUSEABLE | LEV_OPT_CLOSE_ON_FREE,
                                       -1,
                                       (struct sockaddr*)&sin, sizeof(sin));
evconnlistener_set_error_cb(listener, HandleAcceptError);
event_base_dispatch(base);
evconnlistener_free(listener);
event_base_free(base);

I need to support both Linux & Windows. Server process might be killed abruptly at any moment and it should be possible to restart it quickly - hence LEV_OPT_REUSABLE (I think that details are not that important, we are using it in our test setup, server is started&killed by python).

On Linux this behaves as I expect, on Windows I'm getting 10048 (Only one usage of each socket address (protocol/network address/port) is normally permitted.) from evconnlistener_new_bind.

I've learned that SO_REUSEADDR on Unix and on Windows are different and tracked down my problem down to

I assume this is intentional but:

• this behavior could be documented better in

  libevent/include/event2/listener.h

  Lines 69 to 71 in a994a52

  	/** Flag: Indicates that we should disable the timeout (if any) between when
  	* this socket is closed and when we can listen again on the same port. */
  	#define LEV_OPT_REUSEABLE (1u<<3)

  • later I found out that this is mentioned in libevent-book for evutil_make_listen_socket_reuseable description (https://libevent.org/libevent-book/Ref5_evutil.html) although it is not most fortunate place for that.
• perhaps dedicated flag which does set SO_REUSEADDR on Windows could be added so it is possible to use this nice utility function for both platforms - by all means, I'm not an expert, my main complaint here is about documentation - but giving user a choice seems reasonable here.