diff -r 1f466354a85b Lib/sre_compile.py
--- a/Lib/sre_compile.py	Mon Oct 21 09:08:39 2013 +0200
+++ b/Lib/sre_compile.py	Mon Oct 21 13:08:18 2013 +0300
@@ -249,22 +249,18 @@
         return out
     return charset
 
+_CODEBITS = _sre.CODESIZE * 8
 def _mk_bitmap(bits):
-    data = []
-    dataappend = data.append
-    if _sre.CODESIZE == 2:
-        start = (1, 0)
-    else:
-        start = (1, 0)
-    m, v = start
-    for c in bits:
-        if c:
-            v = v + m
-        m = m + m
-        if m > MAXCODE:
-            dataappend(v)
-            m, v = start
-    return data
+    s = bytes(bits).translate(b'0' + b'1' * 255)
+    return [int(s[i: i + _CODEBITS][::-1], 2)
+            for i in range(0, len(s), _CODEBITS)]
+
+def _bytes_to_codes(b):
+    import array
+    a = array.array('I', b)
+    assert a.itemsize == _sre.CODESIZE
+    assert len(a) * a.itemsize == len(b)
+    return a.tolist()
 
 # To represent a big charset, first a bitmap of all characters in the
 # set is constructed. Then, this bitmap is sliced into chunks of 256
@@ -293,10 +289,6 @@
 # bigcharsets.
 
 def _optimize_unicode(charset, fixup):
-    try:
-        import array
-    except ImportError:
-        return charset
     charmap = [0]*65536
     negate = 0
     try:
@@ -326,26 +318,16 @@
     mapping = [0]*256
     block = 0
     data = []
-    for i in range(256):
-        chunk = tuple(charmap[i*256:(i+1)*256])
+    for i in range(0, 65536, 256):
+        chunk = tuple(charmap[i: i + 256])
         new = comps.setdefault(chunk, block)
-        mapping[i] = new
+        mapping[i//256] = new
         if new == block:
-            block = block + 1
-            data = data + _mk_bitmap(chunk)
-    header = [block]
-    if _sre.CODESIZE == 2:
-        code = 'H'
-    else:
-        code = 'I'
-    # Convert block indices to byte array of 256 bytes
-    mapping = array.array('b', mapping).tobytes()
-    # Convert byte array to word array
-    mapping = array.array(code, mapping)
-    assert mapping.itemsize == _sre.CODESIZE
-    assert len(mapping) * mapping.itemsize == 256
-    header = header + mapping.tolist()
-    data[0:0] = header
+            block += 1
+            data.extend(_mk_bitmap(chunk))
+    # Convert block indices to word array
+    mapping = _bytes_to_codes(bytes(mapping))
+    data[0:0] = [block] + mapping
     return [(BIGCHARSET, data)]
 
 def _simple(av):