Re: [bug-gettext] [PATCH] gettext-runtime/intl: Merge from glibc

[Will Newton]

On 13 December 2014 at 03:15, Daiki Ueno <address@hidden> wrote:
> Will Newton <address@hidden> writes:
>
>>>> -#if defined _LIBC_REENTRANT || HAVE_DECL_FGETS_UNLOCKED
>>>> +#if defined _LIBC_REENTRANT || defined HAVE_DECL_FGETS_UNLOCKED
>>>
>>> This is no-no.  On some platforms (e.g. Mac OS X),
>>> HAVE_DECL_FGETS_UNLOCKED will be defined as 0:
>>>
>>>   $ grep 'define HAVE_DECL_FGETS' gettext-runtime/config.h
>>>   #define HAVE_DECL_FGETS_UNLOCKED 0
>>>
>>> What's the rationale behind this change?  According to C99 and C11, the
>>> C preprocessor shall convert undefined identifiers to pp-number 0 before
>>> evaluation.
>>
>> glibc quite recently switched to passing -Wundef to the compiler which
>> requires that #if is only used on defined values in order to catch
>> bugs where somebody forgets or typos the name of a define.
>
> Ah, I just remembered the related discussion on bug-gnulib.
>
>> The usual way of testing config variables that start with HAVE_ is to
>> use #ifdef as they are binary on/off switches. Is there a reason this
>> constant must distinguish between not-defined and zero?
>
> It is an expansion of a standard autoconf macro AC_CHECK_DECLS and is
> a documented behavior:
> http://www.gnu.org/savannah-checkouts/gnu/autoconf/manual/autoconf-2.69/html_node/Generic-Declarations.html#index-AC_005fCHECK_005fDECLS-680
>
> Perhaps we could special case this in intl/, but it would be non-trivial
> and I doubt that it is worth doing just in order to suppress the
> compiler warnings.  So, IMO,
>
>> Converting it to:
>>
>> #if defined _LIBC_REENTRANT || (defined HAVE_DECL_FGETS_UNLOCKED &&
>> HAVE_DECL_FGETS_UNLOCKED)
>>
>> should work but it is a bit ugly.
>
> This is the easier way to work around it (and it seems argp/ already
> does that).

Would you like me to re-submit the patch with this change or are you
happy to fix it up?

Thanks,

-- 
Will Newton
Toolchain Working Group, Linaro