Fix tagtrace.test again

[Ralf Wildenhues]

Automake-1.10 will put in aclocal.m4 a guard that prevents it from being
used with a different Autoconf version from the one that was used by
aclocal to create aclocal.m4.  This means we now have at least three
ways for bogus errors in tests/tagtrace.test:
- at check time, the user's autoconf is a different version than the one
  used to create the Libtool tarball (by the developer); above error and
  exit 63
- the autoconf is new (2.5x) but too old for us (error message
  "Autoconf version ... is required"), and exits 63 (I think some
  intermediate version used a slightly different error message).
- the autoconf version is ancient (2.13, or Debians wrapper for both
  2.13 and 2.5x) and exits 1 with a usage output

OK?

Cheers,
Ralf

        * tests/tagtrace.test: SKIP if `autoconf --trace' exits 63 or 1,
        for various possible (valid) error cases.  Also output stderr,
        to help with analysis.

Index: tests/tagtrace.test
===================================================================
RCS file: /cvsroot/libtool/libtool/tests/tagtrace.test,v
retrieving revision 1.8
diff -u -r1.8 tagtrace.test
--- tests/tagtrace.test 27 Oct 2005 13:17:05 -0000      1.8
+++ tests/tagtrace.test 14 Oct 2006 12:18:28 -0000
@@ -32,10 +32,13 @@
   func_skip "This test requires write access to the source tree"
 fi
 
-if ( cd "$srcdir" && $AUTOCONF --trace 'LT_SUPPORTED_TAG:$1' ) 2>&1 >/dev/null 
|
-     grep "Autoconf version .*is required" >/dev/null; then
-  func_error "This test requires an Autoconf version at least as new"
+( cd "$srcdir" && $AUTOCONF --trace 'LT_SUPPORTED_TAG:$1' ) >/dev/null
+ret=$?
+if test $ret -eq 63 || test $ret -eq 1; then
+  func_error "This test requires the same Autoconf version"
   func_skip  "as the one that was used to bootstrap Libtool"
+elif test $ret -ne 0; then
+  func_fatal_error "\`$AUTOCONF --trace' exited $ret"
 fi
 
 # Abort as soon as something fails.