On 7/2/21 4:59 PM, Torge Matthies wrote:
On 7/2/21 9:33 PM, Zebediah Figura (she/her) wrote:
On 7/2/21 1:06 PM, David Koolhoven wrote:
This makes sure we poll listening non-connection file descriptors even if the event is POLLERR or POLLHUP.
Signed-off-by: David Koolhoven david@koolhoven-home.net
server/sock.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/server/sock.c b/server/sock.c index befa9117c13..ce2f390ec05 100644 --- a/server/sock.c +++ b/server/sock.c @@ -1024,7 +1024,7 @@ static void sock_poll_event( struct fd *fd, int event ) fprintf(stderr, "socket %p select event: %x\n", sock, event); /* we may change event later, remove from loop here */ - if (event & (POLLERR|POLLHUP)) set_fd_events( sock->fd, -1 ); + if (event & (POLLERR|POLLHUP) && sock->state != SOCK_LISTENING) set_fd_events( sock->fd, -1 ); switch (sock->state) {
This doesn't look right. Why do we need to do this?
I only added this because there was no code path with sock->state == SOCK_LISTENING that didn't already lead to a set_fd_events call, which overrides this one, so calling it here is useless. It doesn't fix anything and might add more code complexity than optimizing away one epoll_ctl call is worth.
Personally I'm inclined to say that this code could use less code complexity rather than more (and I doubt the optimization makes a difference anyway.)