Re: g_file_write()

[mortenw gnome org (Morten Welinder)]

Don't expect rename to delete the target file on Win32.  (There goes
the atomicity.)  Checking with g_file_test and unlinking is asking
for a race condition and lost target file in some circumstances.

On unix this type of write will interact in interesting ways with
hardlinks, permissions, file ownership, and non-writable directories.

fdopen is a pita portability-wise, see below.

I'd say go with the simple version you started with and leave the
trouble to the caller.  There are reasons that Emacs has a zillion
settings for what to do.

Morten


-----------------------------------------------------------------------------

AC_CHECK_DECL(fdopen, fdopen_works=yes, fdopen_works=no)
if test $fdopen_works = no ; then
	unset ac_cv_have_decl_fdopen
	CFLAGS="$CFLAGS -D_POSIX_SOURCE"
	AC_MSG_NOTICE([adding -D_POSIX_SOURCE to CFLAGS])
	AC_CHECK_DECL(fdopen, fdopen_works=yes, fdopen_works=no)
	if test $fdopen_works = no ; then
		AC_MSG_ERROR([fdopen is not available])
	fi
fi

# Unfortunately, -D_POSIX_SOURCE turns off struct timeval on Solaris
AC_MSG_CHECKING([whether struct timeval is available])
AC_TRY_COMPILE([#include <sys/time.h>], [struct timeval tv;], struct_timeval_works=yes, struct_timeval_works=no)
AC_MSG_RESULT($struct_timeval_works)

if test $struct_timeval_works = no ; then
	CFLAGS="$CFLAGS -D__EXTENSIONS__"
	AC_MSG_CHECKING([whether struct timeval is available with -D__EXTENSIONS__])
	AC_TRY_COMPILE([#include <sys/time.h>], [struct timeval tv;], struct_timeval_works=yes, struct_timeval_works=no)
	AC_MSG_RESULT($struct_timeval_works)
	if test $struct_timeval_works = no ; then
		AC_MSG_ERROR([struct timeval is not available])
	fi
fi