Undo the previous change and fix the test.

The previous change didn't feel right, so this rolls it back and fixes the
problem at the level of the test, instead.

lib/RPC/XML.pm

@@ -844,7 +844,7 @@ sub as_string
                     ("<member><name>$_</name><value>",
                      $clean{$_},
                      '</value></member>')
-                } (sort keys %clean)),
+                } (keys %clean)),
                 '</struct>';
 }
 
@@ -856,7 +856,7 @@ sub serialize
     my $key;
 
     print {$fh} '<struct>';
-    for (sort keys %{$self})
+    for (keys %{$self})
     {
         ($key = $_) =~ s/$RPC::XML::XMLRE/$RPC::XML::XMLMAP{$1}/ge;
         utf8::downgrade($key);

t/15_serialize.t

@@ -96,7 +96,22 @@ seek $ofh, 0, 0;
 $data = '';
 read $ofh, $data, -s $ofh;
 
-is($data, $faux_res->as_string, 'Fault-response content is correct');
+# There have been some changes to how Perl handles iteration of hash keys.
+# As a result, this test has started failing a lot because of the order of
+# keys when serialized doesn't match the order of keys from as_string(). So
+# to get around this, just compare it to both variations that can occur.
+my $variant1 = '<?xml version="1.0" encoding="us-ascii"?><methodResponse>' .
+    '<fault><value><struct><member><name>faultString</name><value><string>' .
+    'test</string></value></member><member><name>faultCode</name><value>' .
+    '<int>1</int></value></member></struct></value></fault></methodResponse>';
+my $variant2 = '<?xml version="1.0" encoding="us-ascii"?><methodResponse>' .
+    '<fault><value><struct><member><name>faultCode</name><value><int>1</int>' .
+    '</value></member><member><name>faultString</name><value><string>test' .
+    '</string></value></member></struct></value></fault></methodResponse>';
+ok(
+    ($data eq $variant1) || ($data eq $variant2),
+    'Fault-response content is correct'
+);
 
 close $ofh;
 unlink $tmpfile;