Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Sat, 20 Oct 2012 20:11:18 -0400
From: Rich Felker <dalias@...ifal.cx>
To: musl@...ts.openwall.com
Subject: Re: [PATCH 2/4] Add basic sys/cdefs.h found on most unix

On Sun, Oct 21, 2012 at 02:13:47AM +0200, Abdoulaye Walsimou GAYE wrote:
> On 10/21/2012 01:38 AM, Rich Felker wrote:
> >On Sun, Oct 21, 2012 at 01:38:52AM +0200, Abdoulaye Walsimou GAYE wrote:
> >>On 10/21/2012 01:18 AM, Isaac Dunham wrote:
> >>>On Sat, 20 Oct 2012 22:15:43 +0200
> >>>Abdoulaye Walsimou Gaye <awg@...toolkit.org> wrote:
> >>>
> >>>>Signed-off-by: Abdoulaye Walsimou Gaye <awg@...toolkit.org>
> >>>>---
> >>>>  include/sys/cdefs.h |   22 ++++++++++++++++++++++
> >>>>  1 file changed, 22 insertions(+)
> >>>>  create mode 100644 include/sys/cdefs.h
> >>>I'm pretty sure that the last three times sys/cdefs.h was proposed,
> >>>it was rejected.
> >>Unfortunately many packages (wrongly?) use to rely on macros defined there,
> >I've found it's really very few; an equivalent sys/cdefs.h does not
> >exist on most systems. It was never intended for use by applications;
> >it's an internal part of glibc (and perhaps also some BSDs?) used for
> >handling backwardsness like pre-ANSI C compilers (abstracting const
> >away as __const, or abstracting away prototypes with __P()) and
> >optional use of GCC-specific features.
> 
> But applications borrowed from systems internal and ported to others
> systems tend to use these macros (libtirpc, libbsd come in mind).
> Major BSD systems have it (FreeBSD, NetBSD, OpenBSD, PCBSD), it not
> a reason to have it on linux systems but it helps/simplifies packages porting
> from these OS.

As it stands, these libraries/apps won't work anywhere but GNU/Linux
(by "GNU/" I mean glibc-based) or BSD. If the offending code is
removed and replaced with what should be there, they'd be a lot more
portable. So I would not say sys/cdefs.h aids in porting them; I'd say
its presence gives these libs/apps a way to be lazy and
non-portable...

> >For things like 'extern "C"', there's no reason to use sys/cdefs.h;
> >the just writing the code it expands to inline is much more
> >clear/informative and provides better performance as a nice side
> >effect.
> >
> >>sometimes indirectly via <features.h>.
> >I don't see what you mean by this.
> 
> some applications use <features.h> to get macros defined in <sys/cdefs.h>
> as on glibc, eglibc, uClibc <features.h> have a #include <sys/cdefs.h>

Both of these usages are incorrect and could easily be fixed (both are
implementation-internal headers).

Rich

Powered by blists - more mailing lists

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.