+2009-04-10 Bruno Haible <bruno@clisp.org>
+
+ Fix cross-compile guess for utime test.
+ * m4/utime.m4 (AC_FUNC_UTIME_NULL): Add definition from newest unstable
+ autoconf.
+ * doc/posix-functions/utime.texi: Give more precisions.
+ Reported by Jan <ipif@ymail.com>.
+
2009-04-09 Kamil Dudka <kdudka@redhat.com>
filevercmp: correct today's change
Portability problems fixed by Gnulib:
@itemize
@item
-On some old platforms, @code{utime (file, NULL)} fails to set the file's
-timestamp to the current time.
+On some old platforms (Sequent), @code{utime (file, NULL)} fails to set the
+file's timestamp to the current time.
@end itemize
Portability problems not fixed by Gnulib:
-# serial 8
+# serial 9
dnl From Jim Meyering
dnl Replace the utime function on systems that need it.
AC_REQUIRE([gl_CHECK_TYPE_STRUCT_UTIMBUF])
gl_FUNC_UTIMES_NULL
])
+
+# Use the definition of AC_FUNC_UTIME_NULL from autoconf 2.64 or newer.
+# Remove this macro when we can assume autoconf >= 2.64.
+m4_version_prereq([2.64], [], [
+AC_DEFUN([AC_FUNC_UTIME_NULL],
+[AC_CHECK_HEADERS_ONCE([utime.h])
+AC_CACHE_CHECK([whether utime accepts a null argument], [ac_cv_func_utime_null],
+[rm -f conftest.data; >conftest.data
+# Sequent interprets utime(file, 0) to mean use start of epoch. Wrong.
+AC_RUN_IFELSE([AC_LANG_PROGRAM([AC_INCLUDES_DEFAULT
+ #ifdef HAVE_UTIME_H
+ # include <utime.h>
+ #endif],
+[[struct stat s, t;
+ return ! (stat ("conftest.data", &s) == 0
+ && utime ("conftest.data", 0) == 0
+ && stat ("conftest.data", &t) == 0
+ && t.st_mtime >= s.st_mtime
+ && t.st_mtime - s.st_mtime < 120);]])],
+ ac_cv_func_utime_null=yes,
+ ac_cv_func_utime_null=no,
+ ac_cv_func_utime_null='guessing yes')])
+if test "x$ac_cv_func_utime_null" != xno; then
+ ac_cv_func_utime_null=yes
+ AC_DEFINE([HAVE_UTIME_NULL], [1],
+ [Define to 1 if `utime(file, NULL)' sets file's timestamp to the
+ present.])
+fi
+rm -f conftest.data
+])
+])