#5057: the peepholer no longer optimizes subscription on unicode literals (e.g.…

#5057: the peepholer no longer optimizes subscription on unicode literals (e.g. u"foo"[0]) in order to produce compatible pyc files between narrow and wide builds.

Lib/test/test_peepholer.py

@@ -196,21 +196,22 @@ class TestTranforms(unittest.TestCase):
         self.assertIn('(1000)', asm)
 
     def test_binary_subscr_on_unicode(self):
-        # valid code get optimized
+        # unicode strings don't get optimized
         asm = dis_single('"foo"[0]')
-        self.assertIn("('f')", asm)
-        self.assertNotIn('BINARY_SUBSCR', asm)
+        self.assertNotIn("('f')", asm)
+        self.assertIn('BINARY_SUBSCR', asm)
         asm = dis_single('"\u0061\uffff"[1]')
-        self.assertIn("('\\uffff')", asm)
-        self.assertNotIn('BINARY_SUBSCR', asm)
+        self.assertNotIn("('\\uffff')", asm)
+        self.assertIn('BINARY_SUBSCR', asm)
 
-        # invalid code doesn't get optimized
         # out of range
         asm = dis_single('"fuu"[10]')
         self.assertIn('BINARY_SUBSCR', asm)
         # non-BMP char (see #5057)
         asm = dis_single('"\U00012345"[0]')
         self.assertIn('BINARY_SUBSCR', asm)
+        asm = dis_single('"\U00012345abcdef"[3]')
+        self.assertIn('BINARY_SUBSCR', asm)
 
 
     def test_folding_of_unaryops_on_constants(self):

Misc/NEWS

@@ -10,6 +10,10 @@ What's New in Python 3.2.4
 Core and Builtins
 -----------------
 
+- Issue #5057: the peepholer no longer optimizes subscription on unicode
+  literals (e.g. u'foo'[0]) in order to produce compatible pyc files between
+  narrow and wide builds.
+
 - Issue #16402: When slicing a range, fix shadowing of exceptions from
   __index__.
 

Python/peephole.c

@@ -132,25 +132,14 @@ fold_binops_on_constants(unsigned char *codestr, PyObject *consts)
             newconst = PyNumber_Subtract(v, w);
             break;
         case BINARY_SUBSCR:
-            newconst = PyObject_GetItem(v, w);
             /* #5057: if v is unicode, there might be differences between
-               wide and narrow builds in cases like '\U00012345'[0].
-               Wide builds will return a non-BMP char, whereas narrow builds
-               will return a surrogate.  In both the cases skip the
-               optimization in order to produce compatible pycs.
-             */
-            if (newconst != NULL &&
-                PyUnicode_Check(v) && PyUnicode_Check(newconst)) {
-                Py_UNICODE ch = PyUnicode_AS_UNICODE(newconst)[0];
-#ifdef Py_UNICODE_WIDE
-                if (ch > 0xFFFF) {
-#else
-                if (ch >= 0xD800 && ch <= 0xDFFF) {
-#endif
-                    Py_DECREF(newconst);
-                    return 0;
-                }
-            }
+               wide and narrow builds in cases like '\U00012345'[0] or
+               '\U00012345abcdef'[3], so it's better to skip the optimization
+               in order to produce compatible pycs.
+            */
+            if (PyUnicode_Check(v))
+                return 0;
+            newconst = PyObject_GetItem(v, w);
             break;
         case BINARY_LSHIFT:
             newconst = PyNumber_Lshift(v, w);