Re: [PATCH]_BSD_SOURCE for musl, take 2.

[Rich Felker <dalias@aerifal.cx>]

On Mon, May 21, 2012 at 11:31:58AM -0700, Isaac Dunham wrote:
> Here's a second verion of the patch, with all the issues below
> addressed.

Looks good. A few more things, mostly just questions to make sure
stuff is ok..

> I'm still declaring bsd_signal if _BSD_SOURCE is defined, because it
> uses the signature of BSD signal().

Is there a reason for this? bsd_signal was a legacy SUSv3 function
removed in SUSv4/POSIX-2008, which was originally added mainly so that
programs written for BSD signal semantics could do:

#define signal bsd_signal

on non-BSD systems and still work. On glibc and musl it's utterly
useless since signal, by default, has BSD semantics. (POSIX allows
either and only BSD is sane.)

I doubt any code for BSD would ever be calling bsd_signal, as the
whole purpose of it is to provide an alternate function with BSD
behavior on non-BSD systems.

> diff --git a/include/netinet/tcp.h b/include/netinet/tcp.h
> index c8a1a4b..797ce68 100644
> --- a/include/netinet/tcp.h
> +++ b/include/netinet/tcp.h
> @@ -2,5 +2,22 @@
>  #define _NETINET_TCP_H
>  
>  #define TCP_NODELAY 1
> +#if defined(_GNU_SOURCE) || defined(_BSD_SOURCE)
> +#include <sys/types.h>
> +#include <sys/socket.h>
> +#define TCP_MAXSEG	 2
> +#define TCP_CORK	 3
> +#define TCP_KEEPIDLE	 4
> [...]

TCP_* is in the reserved namespace for this header, so I think it
would be reasonable to define all the macros unconditionally. What
about the two includes, though? If they're needed to make GNU/BSD
programs work, we should still keep the check just for them.

> diff --git a/include/setjmp.h b/include/setjmp.h
> index b024c44..88fb950 100644
> --- a/include/setjmp.h
> +++ b/include/setjmp.h
> @@ -9,7 +9,8 @@ extern "C" {
>  
>  
>  #if defined(_POSIX_SOURCE) || defined(_POSIX_C_SOURCE) \
> - || defined(_XOPEN_SOURCE) || defined(_GNU_SOURCE)
> + || defined(_XOPEN_SOURCE) || defined(_GNU_SOURCE) \
> + || defined(_BSD_SOURCE)
>  typedef unsigned long sigjmp_buf[(128+sizeof(jmp_buf))/sizeof(long)];
>  #ifdef _GNU_SOURCE
>  #define jmp_buf sigjmp_buf
   ^^^^^^^^^^^^^^^^^^^^^^^^^^
Just reminding myself this define is at best useless and at worst
harmful (big bloat). But changing it is separate; I'll do it after the
BSD commit.

> +#if defined(_GNU_SOURCE) || defined(_BSD_SOURCE)
> +void (*bsd_signal(int, void (*)(int)))(int);
> +#endif

See comments at the top of this email. Unless there's a really good
reason to declare this with BSD_SOURCE I'd like to leave it just under
_GNU_SOURCE.

> diff --git a/include/unistd.h b/include/unistd.h
> index b1a84d7..0112276 100644
> --- a/include/unistd.h
> +++ b/include/unistd.h
> @@ -16,6 +16,12 @@ extern "C" {
>  #define SEEK_CUR 1
>  #define SEEK_END 2
>  
> +#if defined(_BSD_SOURCE) && !defined(L_SET)
> +#define L_SET	SEEK_SET
> +#define L_INCR	SEEK_CUR
> +#define L_XTND	SEEK_END

Is L_SET defined somewhere else?

> +#ifdef _GNU_SOURCE
> +pid_t forkall(void);

Also mainly a reminder to me: forkall is actually no longer supported
(and was never in glibc) and should just be removed, but I'll do that
in a separate patch too. (The semantics the function intends to meet
are just logically contradictory/impossible).

Rich