Libvirt CI recently started running "make check" on our FreeBSD 10 & 11 hosts, and we see frequent failure of the test-poll unit test in gnulib IIUC, gnulib is not actually building a replacement poll() function on FreeBSD, it is merely running the gnulib test suite against the FreeBSD system impl of poll() and hitting this behaviour.
$ ./gnulib/tests/test-poll Unconnected socket test... passed Connected sockets test... failed (expecting POLLHUP after shutdown) General socket test with fork... failed (expecting POLLHUP after shutdown) Pipe test... passed Looking at the first failure in test_socket_pair method. It sets up a listener socket, connects a client, accepts the client and then closes the remote end. It expects the server's client socket to thus show POLLHUP or POLLERR. When it fails, the poll() is in fact still showing POLLOUT. If you put a sleep between the close () and poll() calls, it will succeed. So, IIUC, the test is racing with the BSD kernel's handling of socket close - the test can't assume that just because the remote end of the client has been closed, that poll() will immediately show POLLHUP|ERR. Anyone have ideas on how to make this test more reliable and not depend on the kernel synchronizing the close() state with poll() results immediately ? Regards, Daniel -- |: https://berrange.com -o- https://www.flickr.com/photos/dberrange :| |: https://libvirt.org -o- https://fstop138.berrange.com :| |: https://entangle-photo.org -o- https://www.instagram.com/dberrange :|
