m4/openat.m4 robustness

[Bruno Haible]

Hi Jim,

The macros in m4/openat.m4 could be made a little more reliable:
  - Before running a test program that invokes fchownat(), make sure that
    glibc will declare fchownat().
  - Ensure that REPLACE_FCHOWNAT at is not set to 0 accidentally after it
    has already been determined that its correct value should be 1.

In my tests, I didn't see the wrong macro ordering occur actually,
but the bugs could surface if macro invocations get rearranged, or
if fchownat is moved to a module of its own, or similar.

OK to apply?


2010-11-13  Bruno Haible  <address@hidden>

        openat: Ensure autoconf macro ordering.
        * m4/openat.m4 (gl_FUNC_FCHOWNAT_DEREF_BUG): Require
        gl_USE_SYSTEM_EXTENSIONS.
        (gl_FUNC_FCHOWNAT): Require gl_UNISTD_H_DEFAULTS.

--- m4/openat.m4.orig   Sat Nov 13 13:58:48 2010
+++ m4/openat.m4        Sat Nov 13 13:56:13 2010
@@ -63,6 +63,9 @@
 # gl_FUNC_FCHOWNAT_DEREF_BUG([ACTION-IF-BUGGY[, ACTION-IF-NOT_BUGGY]])
 AC_DEFUN([gl_FUNC_FCHOWNAT_DEREF_BUG],
 [
+  dnl Persuade glibc <unistd.h> to declare fchownat().
+  AC_REQUIRE([gl_USE_SYSTEM_EXTENSIONS])
+
   AC_CACHE_CHECK([whether fchownat works with AT_SYMLINK_NOFOLLOW],
     gl_cv_func_fchownat_nofollow_works,
     [
@@ -104,6 +107,7 @@
 # Also use the replacement function if fchownat is simply not available.
 AC_DEFUN([gl_FUNC_FCHOWNAT],
 [
+  AC_REQUIRE([gl_UNISTD_H_DEFAULTS])
   AC_REQUIRE([gl_FUNC_CHOWN])
   AC_CHECK_FUNC([fchownat],
     [gl_FUNC_FCHOWNAT_DEREF_BUG(