Hi Eric,
> We were blindly defining ssize_t to int. On mingw64, this is the
> correct size, but the wrong rank, which leads gcc to issue warnings
> for %zd printf directives.
>
> * m4/ssize_t.m4 (gt_TYPE_SSIZE_T): Match rank of size_t.
This patch will not help to fix Daniel Berrange's problem. In mingw
and mingw64, ssize_t is defined by <sys/types.h>, and gnulib will not
override it.
Test program:
==================== foo.c =================
#include <sys/types.h>
size_t a;
ssize_t b;
============================================
With mingw of 2009:
$ gcc -mno-cygwin -c foo.c
$ gcc -mno-cygwin -E foo.c | grep size_t
typedef unsigned int size_t;
typedef long _ssize_t;
typedef _ssize_t ssize_t;
size_t a;
ssize_t b;
With mingw of 2011:
$ i686-pc-mingw32-gcc -c foo.c
$ i686-pc-mingw32-gcc -E foo.c | grep size_t
typedef unsigned int size_t;
typedef int _ssize_t;
typedef _ssize_t ssize_t;
size_t a;
ssize_t b;
With mingw64 of 2011:
$ i686-w64-mingw32-gcc -c foo.c
$ i686-w64-mingw32-gcc -E foo.c | grep size_t
typedef unsigned int size_t;
typedef int ssize_t;
size_t a;
ssize_t b;
This is consistent with what doc/posix-headers/sys_types.texi says:
@item
The type @code{ssize_t} is not defined on some platforms:
MSVC 9.
> Bruno, is this okay to apply?
No. Even if it would help, there would be a few nits:
> + AC_CACHE_CHECK([for rank of size_t], [gt_cv_size_t_rank],
Better say "integer conversion rank", since that's the term ISO C 99
uses (section 6.3.1.1).
> + [AC_COMPILE_IFELSE(
> + [AC_LANG_PROGRAM(
> + [[#include <sys/types.h>
> + #ifdef __cplusplus
Intentation problem.
> + extern "C" {
> + #endif
> + int foo(unsigned long bar);
> + int foo(size_t bar);
You cannot look at size_t to determine what ssize_t should be: On MSVC 9,
<sys/types.h> lacks both size_t and ssize_t.
> + ]])],
> + [gt_cv_size_t_rank=long], [gt_cv_size_t_rank=int])])
Another indentation problem.
In summary, I would not do something about it in gnulib. The proper place
for setting up size_t and ssize_t are GCC and <sys/types.h>. If gnulib
would override the definition, it would have side effects on C++ overloading.
For my feeling, that's not worth it. Simply ignore the GCC printf argument
size warnings on mingw. Look at them only on glibc systems.
Bruno
