Werror in <sys/socket.h>

[Richard Henderson <rth at twiddle dot net>]

Miraculously, alphaev67 builds with gcc 4.9.2 without Werror.

Perusing make check, however, I see

> In file included from ../inet/netinet/in.h:23:0,
>                  from ../include/netinet/in.h:3,
>                  from ../resolv/resolv.h:57,
>                  from ../include/resolv.h:12,
>                  from ../nptl/descr.h:39,
>                  from ../sysdeps/alpha/nptl/tls.h:48,
>                  from ../sysdeps/unix/sysv/linux/alpha/sysdep.h:30,
>                  from ../sysdeps/unix/sysv/linux/lowlevellock-futex.h:22,
>                  from ../sysdeps/nptl/lowlevellock.h:23,
>                  from ../sysdeps/nptl/bits/stdio-lock.h:23,
>                  from ../libio/libio.h:149,
>                  from ../libio/stdio.h:74,
>                  from scanf15.c:10:
> ../include/sys/socket.h:95:24: error: âstruct mmsghdrâ declared inside parameter
>  list [-Werror]
>                         unsigned int __vlen, int __flags);
>                         ^
> ../include/sys/socket.h:95:24: error: its scope is only this definition or decla
> ration, which is probably not what you want [-Werror]
> cc1: all warnings being treated as errors
> make[2]: *** [/home/rth/work/libc/bld/stdio-common/scanf15.o] Error 1

I'm not sure what the proper resolution for this should be.

Normally we have __USE_GNU defined, and so pick up the definition of mmsghdr
from <socket/sys/socket.h>.  But this test explicitly uses _XOPEN_SOURCE, so we
don't have that.

Should <include/sys/socket.h> mirror <socket/sys/socket.h> in suppressing the
definition of __sendmmsg if __USE_GNU is not defined?  Or should we just add a
forward declaration of the structure on the preceding line?


r~