[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [PATCH] bypass broken inline strtod() definition in <stdlib.h> on Mi

From: Bruno Haible
Subject: Re: [PATCH] bypass broken inline strtod() definition in <stdlib.h> on Mingw
Date: Fri, 11 Jun 2010 11:02:23 +0200
User-agent: KMail/1.9.9

Hi Ben,

> Cross-compiling lib/strtod.c failed on Mingw, as reported at
> http://savannah.gnu.org/bugs/?29965:
>     strtod.c:37: error: redefinition of 'strtod'
>     /usr/i686-pc-mingw32/sys-root/mingw/include/stdlib.h:319: note:
>     previous definition of 'strtod' was here 
> The problem was the section of stdlib.h cited in the above
> message:
>     double __cdecl __MINGW_NOTHROW __strtod (const char*, char**);
>     #ifdef __cplusplus
>     /* We require a function with external linkage. */
>     #else
>     static
>     #endif /* Not __cplusplus */
>     __inline__ double __cdecl __MINGW_NOTHROW
>     strtod (const char* __restrict__ __nptr, char** __restrict__ __endptr)
>     {
>       return __strtod(__nptr, __endptr);
>     }
>     float __cdecl __MINGW_NOTHROW strtof (const char * __restrict__, char ** 
> __restrict__);
>     long double __cdecl __MINGW_NOTHROW strtold (const char * __restrict__, 
> char ** __restrict__);
> As you can see, this defines an inline version of strtod() that
> conflicts with the out-of-line version in lib/strtod.c.
> Is there an idiomatic solution for this kind of problem?

Why does it conflict? gnulib's <stdlib.h> replacement already contains the
idiomatic solution. As you can see from m4/strtod.m4, there are three cases:
  - function strtod ok
  - function strtod does not exist
    => HAVE_STRTOD=0, lib/strtod.c gets compiled
  - function strtod does not obey C99
    => REPLACE_STRTOD=1, lib/strtod.c gets compiled

You are probably in the 3rd case? (Please look into your config.status to

The stdlib.in.h contains exactly the right idiom for these 3 cases:

   /* Parse a double from STRING, updating ENDP if appropriate.  */
  #  if !(defined __cplusplus && defined GNULIB_NAMESPACE)
  #   define strtod rpl_strtod
  #  endif
  _GL_FUNCDECL_RPL (strtod, double, (const char *str, char **endp)
                                    _GL_ARG_NONNULL ((1)));
  _GL_CXXALIAS_RPL (strtod, double, (const char *str, char **endp));
  # else
  #  if address@hidden@
  _GL_FUNCDECL_SYS (strtod, double, (const char *str, char **endp)
                                    _GL_ARG_NONNULL ((1)));
  #  endif
  _GL_CXXALIAS_SYS (strtod, double, (const char *str, char **endp));
  # endif
  _GL_CXXALIASWARN (strtod);

In the 3rd case, you should see the
   #define strtod rpl_strtod

and this #define should avoid a conflict between the function in the system
header and the function in gnulib.

> I'm appending my fix, which I have not yet pushed out.

You need first to analyze why the existing gnulib idiom did not work in this
situation. (Take a look at config.status and config.log. Look at the
"gcc -E -dD" output. And so on.)


reply via email to

[Prev in Thread] Current Thread [Next in Thread]