Re: utmpxname() but no prototype?

[Rich Felker <dalias@libc.org>]

On Fri, Feb 12, 2016 at 12:28:53AM -0800, Andre McCurdy wrote:
> Hi all,
> 
> The lxc configure script uses AC_CHECK_FUNCS to test for utmpxname()
> support. From the comments it looks like this check was added
> specifically for compatibility with musl:
> 
>   https://github.com/lxc/lxc/commit/8b6d8b712b867ab352598ed4b73e80e54a8c915a
> 
> Up until recently, this worked as expected: the configure script
> correctly detected that musl did not provide utmpxname().
> 
> However, recently musl has gained a utmpxname() stub:
> 
>   http://git.musl-libc.org/cgit/musl/commit/?id=378f8cb5222b63e4f8532c757ce54e4074567e1f
> 
> but without also gaining a corresponding prototype in utmpx.h.
> 
> This causes a new problem when building lxc: the configure script now
> detects that utmpxname() is provided but the build then fails because
> there's no prototype for it:
> 
>  | ../../../lxc-1.0.7/src/lxc/lxcutmp.c: In function 'utmp_get_runlevel':
>  | ../../../lxc-1.0.7/src/lxc/lxcutmp.c:256:30: error: implicit
> declaration of function 'utmpxname'
> [-Werror=implicit-function-declaration]
>  |   if (!access(path, F_OK) && !utmpxname(path))
>  |                               ^
> 
> Passing "ac_cv_func_utmpxname=no" to the lxc configure script is a
> workaround but I'm wondering what the real solution should be. Should
> utmpx.h be providing:
> 
>   #define utmpxname(x) (-1)
> 
> in the same way that utmp.h provides:
> 
>   #define utmpname(x) (-1)
> 
> ?

No, there should be a prototype for it (under the appropriate feature
test conditions); this is simply an oversight. I'll add it. Thanks for
the report.

Rich