This is the mail archive of the
mailing list for the glibc project.
Re: static_assert in c89 and c99
- From: Rich Felker <dalias at libc dot org>
- To: libc-alpha at sourceware dot org
- Date: Wed, 4 Feb 2015 09:35:35 -0500
- Subject: Re: static_assert in c89 and c99
- Authentication-results: sourceware.org; auth=none
- References: <87wq3x27di dot fsf at black dot gullik dot net>
On Wed, Feb 04, 2015 at 01:23:05PM +0100, Lars Gullik BjÃnnes wrote:
> In our sources in pre c11 we add our own assert.h and use #include_next
> to get the proper one after having added a couple of things of our own.
> One of those things is a static_assert defined when not compiling with
> -std=c11, if c11 we want the one provided by glibcs assert.h.
> test snippet:
> #include <assert.h>
> #ifdef static_assert
> #error static_assert defined
> gcc -D_GNU_SOURCE -std=c89 -c snippet.c
> gcc -D_GNU_SOURCE -std=c99 -c snippet.c
> gcc -D_GNU_SOURCE -std=c11 -c snippet.c
> all error out, where only c11 variant is expected to do so.
> It seems that the addition of "# define _ISOC11_SOURCE 1" (features.h)
> to what gets defined by _GNU_SOURCE is causing this.
Defining _GNU_SOUCE gives glibc permission to expose anything and
everything. If you don't want this, don't define it. The whole point
of _GNU_SOURCE is to free glibc from namespace constraints.
> In assert.h static_assert is always defined when __USE_ISOC11 is defined
> and that is unconditionally defined by _ISOC11_SOURCE. For a library
> feature that would be ok, but since static_assert is using
> _Static_assert its usability is really controlled by the compiler and
> I think the static_assert define must be guarded by a check that
> -std=c11 really is in effect.
If glibc is providing static_assert using _Static_assert on compilers
which don't implement the latter, that should probably be fixed. It
could be done using _Static_assert conditionally on __GNUC__>=x ||
__STDC_VERSION__>=C11, with a fallback to an old method like the dummy
type declaration with possibly-invalid array size.