Re: filemode with C++

[Jim Meyering]

Bruno Haible wrote:

> Hi John,
>
>> I'd like to use the filemode module in a C++ file but found that it
>> does not have extern "C" magic.  I started to submit a patch, but
>> stopped when I saw that filemode.h has
>>
>>   # if HAVE_DECL_STRMODE
>>   #  include <string.h> /* FreeBSD, OpenBSD */
>>   #  include <unistd.h> /* NetBSD */
>>   # else
>>   void strmode (mode_t mode, char *str);
>>   # endif
>>
>>   void filemodestring (struct stat const *statp, char *str);
>>
>> and remembered there is a recommendation in the gnulib manual to not
>> surround system header files with an extern "C" block.
>>
>> What's the best way to get declare these two functions extern "C" and
>> be consistent with gnulib conventions?
>
> The best way is to first include all necessary system header files, then,
> in an extern "C" block, do all variable or function declarations.
>
> It is OK to duplicate a #if condition for this purpose.
>
>> Is it safe to assume that if HAVE_DECL_STRMODE is true, strmode will
>> be declared extern "C"?
>
> Yes. I checked
>
> MacOS X 10.3 <string.h>
> FreeBSD 6.4 <string.h>
> OpenBSD 3.8 <string.h>
> NetBSD 3.0 <unistd.h>
>
> and they all look like this:
>
> __BEGIN_DECLS
> ...
> void   strmode(mode_t, char *);
> ...
> __END_DECLS
>
> In general, nowadays, it's rare to find a system function that is _not_
> properly declared inside extern "C" for C++. (Rare example: tcgetsid()
> on HP-UX 11.00.)
>
> Here's a proposed patch. In passing, this patch adds an 'extern' modifier',
> which is redundant for functions, but nice to have
>   - for consistency between variable and function declarations,
>   - for consistency with the style of the glibc header files,
>   - for determining the desired set of exported symbols of a shared library
>     from its include files.
>
> Jim, OK to commit?
>
>
> 2011-01-14  Bruno Haible  <address@hidden>
>
>       filemode: Make function declarations usable in C++ mode.
>       * lib/filemode.h: Enclose function declarations in extern "C" block.
>       Reported by John W. Eaton <address@hidden>.

Yes, that looks fine.  Thanks, Bruno.