Re: [3/11] Native MSVC support (msvc-cwrapper)

[Charles Wilson]

Peter Rosin wrote:
> On Thu, Jul 19, 2007 at 05:47:36PM -0400, Charles Wilson wrote:
> The cwrapper has two targets (cygwin, mingw), I'm adding a third (msvc).
> Both the previous targets has stat as stat, cygwin doesn't have _stat
> and msvc doesn't have stat (unless you link with OLDNAMES.LIB).
>
> So, going the other way around breaks cygwin (IIRC). 

How can it break cygwin?

+#ifdef _MSC_VER
+# define S_IXUSR _S_IEXEC
+# define _stat stat
+# ifndef _INTPTR_T_DEFINED
+#  define intptr_t int
+# endif
+#endif

cygwin does not #define _MSC_VER (nor does mingw, AFAIK).  So, the only 
thing that matters here, is:

(1) the cwrapper code invokes stat().
(2) on cygwin this works.
(3) on mingw this works, because -lmoldnames is linked in via libgcc's 
dependencies in the spec file. At least, for now.
(4) on msvc this is broken, because the only thing exported by 
msvcrt.dll is _stat.

Therefore, on msvc only, we want the cwrapper to call _stat() instead of 
stat().  Thus:

#ifdef _MSC_VER
# define stat _stat
#endif

> But I guess mingw
> could drop stat at some point as well (and require -loldnames) in order
> to match the MS libc. I think mingw has _stat already.

Yes, mingw has _stat via the libmsvcrt.a import library. But right now, 
it automatically gets stat() via libmoldnames, so there's no need to 
change anything.

> Should I commit as is?

Since the only line in dispute is guarded by _MSC_VER, I believe that is 
up to you.  Whether that line is right or wrong, it won't break mingw 
nor cygwin.  But you should test under both MSVS-2003 and MSVS-2005, to 
make sure you've got that particular line correct.

--
Chuck