> From: Bruno Haible <br...@clisp.org>
> Cc: Tim Rühsen <tim.rueh...@gmx.de>
> Date: Sat, 12 May 2018 14:31:59 +0200
>
> > This is because configure-time test for _fseeki64 succeeds, but the
> > prototype of _fseeki64 is not visible unless the condition
> >
> > _WIN32_WINNT >= 0x0600
> >
> > holds
>
> According to
> https://sourceforge.net/p/mingw/mingw-org-wsl/ci/5.1-trunk/tree/mingwrt/include/stdio.h#l743
> the condition should be
> _WIN32_WINNT >= _WIN32_WINNT_VISTA || __MSVCRT_VERSION__ >= __MSVCR80_DLL
> I guess...
Yes. They both express the same intent, in slightly different ways
(which have subtly different effects, but I don't think we should
muddy the waters by bringing them up here). The intent is to target
MSVCRT versions that came with Vista and later versions.
> Anyway, the issue is that the "old" mingw defines fseeko64 instead of
> _fseeki64, and the gnulib code attempts to distinguish the two cases:
>
> # if HAVE__FSEEKI64 /* msvc, mingw64 */
> # define fseeko _fseeki64
> # else /* mingw */
> # define fseeko fseeko64
> # endif
>
> But the distinction between mingw-w64 and mingw.org is broken.
>
> Do you know how to distinguish the two, just by preprocessor defines,
> without any autoconf test?
AFAIK, it's tricky, because MinGW64 also defines __MINGW32__, and
doesn't have any macro specific to it that is defined by the compiler
itself. In Emacs we use this ugly hack to resolve the problem:
/* MinGW-w64 gcc does not automatically define a macro for
differentiating it fom MinGW gcc. We need to test the presence of
__MINGW64_VERSION_MAJOR in _mingw.h: */
#ifdef __MINGW32__
# include <_mingw.h>
# ifdef __MINGW64_VERSION_MAJOR
# define MINGW_W64
# endif
#endif
and then test MINGW_W64 elsewhere.
