bug#12116: merge from gnulib for extern-inline

[Eli Zaretskii <eliz@gnu.org>]

> Date: Thu, 02 Aug 2012 11:02:59 -0700
> From: Paul Eggert <eggert@cs.ucla.edu>
> CC: 12116@debbugs.gnu.org, lekktu@gmail.com
> 
>                              e x p a n s i o n   i n :
>    macro           C99            older compilers   ancient compilers
>    INLINE          inline         static inline     static
>    EXTERN_INLINE   extern inline  static inline     static
> 
> and the Emacs include file charset.h does this:
> 
>    #ifndef CHARSET_INLINE
>    # define CHARSET_INLINE INLINE
>    #endif
> 
>    CHARSET_INLINE void
>    set_charset_attr (struct charset *charset, enum charset_attr_index idx,
>                      Lisp_Object val)
>    {
>      ASET (CHARSET_ATTRIBUTES (charset), idx, val);
>    }
> 
> so in C99 most users of charset.h see the function as being
> 'inline void'.  charset.c does this before including charset.h:
> 
>    #define CHARSET_INLINE EXTERN_INLINE
> 
> so that when it's compiled, the function is 'extern inline void'.
> In pre-C99 compilers the function is 'static inline' or (if ancient)
> plain 'static'.  Similarly for each .h file that uses inline functions.

OK, but what's the story with the addition of stat-time.c,
utimespec.c, and u64.c, which just include the respective headers?
Where and how are these used, and for what purposes?

I understood the extern inline stuff, I just don't get why are these
*.c files introduced.