On 8/20/19 11:37 AM, Eric Blake wrote:
> Ideally, we would improve our replacement <sys/socket.h> to define a
> replacement SOCK_NONBLOCK on all platforms, and teach socket() to
> honor it as well; but that's a bigger task.  In the meantime, if the
> platform already has SOCK_NONBLOCK, we should honor it when doing a
> fallback.
> 

> 
> +#if SOCK_CLOEXEC
> +  if (flags & SOCK_NONBLOCK)
> +    {
> +      int fcntl_flags;
> +
> +      if ((fcntl_flags = fcntl (fd, F_GETFL, 0)) < 0
> +          || fcntl (fd, F_SETFL, fcntl_flags | O_NONBLOCK) == -1)

Hmm, I guess it would be better to make the accept4 module depend on
nonblocking, and call set_nonblocking() instead of open-coding the
Unix-only solution (but this works for now as long as mingw lacks
SOCK_NONBLOCK natively, and as long as we don't have <sys/socket.h>
declaring a replacement SOCK_NONBLOCK)...



-- 
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3226
Virtualization:  qemu.org | libvirt.org