Re: [PATCH] Assume that accept4 is available

[Joseph Myers <joseph at codesourcery dot com>]

On Fri, 14 Apr 2017, Florian Weimer wrote:

> diff --git a/sysdeps/unix/sysv/linux/kernel-features.h b/sysdeps/unix/sysv/linux/kernel-features.h
> index fd936c5..b981466 100644
> --- a/sysdeps/unix/sysv/linux/kernel-features.h
> +++ b/sysdeps/unix/sysv/linux/kernel-features.h
> @@ -75,18 +75,11 @@
>     architectures using a separate syscall rather than socketcall that
>     syscall was only added later, and some architectures first had
>     socketcall support then a separate syscall.  Define
> -   __ASSUME_ACCEPT4_SOCKETCALL if glibc uses socketcall on this
> -   architecture and accept4 is available through socketcall,
>     __ASSUME_ACCEPT4_SYSCALL if it is available through a separate
> -   syscall, __ASSUME_ACCEPT4_SYSCALL_WITH_SOCKETCALL if it became
> +   syscall, and __ASSUME_ACCEPT4_SYSCALL_WITH_SOCKETCALL if it became
>     available through a separate syscall at the same time as through
> -   socketcall, and __ASSUME_ACCEPT4 if the accept4 function is known
> -   to work.  */
> -#ifdef __ASSUME_SOCKETCALL
> -# define __ASSUME_ACCEPT4_SOCKETCALL	1
> -#endif
> +   socketcall.  */
>  #define __ASSUME_ACCEPT4_SYSCALL	1
> -#define __ASSUME_ACCEPT4	1
>  
>  /* Support for the FUTEX_CLOCK_REALTIME flag was added in 2.6.29.  */
>  #define __ASSUME_FUTEX_CLOCK_REALTIME	1

By removing the description and definition of __ASSUME_ACCEPT4_SOCKETCALL, 
this loses the information about what __ASSUME_RECVMMSG_SOCKETCALL and 
__ASSUME_SENDMMSG_SOCKETCALL mean, since the comments on the macros for 
recvmmsg and sendmmsg define them by reference to the corresponding macros 
for accept4.

-- 
Joseph S. Myers
joseph@codesourcery.com