The simplest possible fix for the regression in bug 12752 by encoding unicodes
to 8-bit strings.
diff --git a/Lib/locale.py b/Lib/locale.py
index 166538d..0c4d652 100644
--- a/Lib/locale.py
+++ b/Lib/locale.py
@@ -355,6 +355,8 @@
 
     """
     # Normalize the locale name and extract the encoding
+    if isinstance(localename, unicode):
+        localename = localename.encode('ascii')
     fullname = localename.translate(_ascii_lower_map)
     if ':' in fullname:
         # ':' is sometimes used as encoding delimiter.