Issue #11244: The peephole optimizer is now able to constant-fold

arbitrarily complex expressions.  This also fixes a 3.2 regression where
operations involving negative numbers were not constant-folded.

Lib/test/test_peepholer.py

@@ -8,7 +8,9 @@ def disassemble(func):
     f = StringIO()
     tmp = sys.stdout
     sys.stdout = f
+    try:
         dis.dis(func)
+    finally:
         sys.stdout = tmp
     result = f.getvalue()
     f.close()
@@ -99,6 +101,12 @@ class TestTranforms(unittest.TestCase):
             self.assertIn(elem, asm)
             self.assertNotIn('BUILD_TUPLE', asm)
 
+        # Long tuples should be folded too.
+        asm = dis_single(repr(tuple(range(10000))))
+        # One LOAD_CONST for the tuple, one for the None return value
+        self.assertEqual(asm.count('LOAD_CONST'), 2)
+        self.assertNotIn('BUILD_TUPLE', asm)
+
         # Bug 1053819:  Tuple of constants misidentified when presented with:
         # . . . opcode_with_arg 100   unary_opcode   BUILD_TUPLE 1  . . .
         # The following would segfault upon compilation
@@ -267,6 +275,25 @@ class TestTranforms(unittest.TestCase):
         asm = disassemble(f)
         self.assertNotIn('BINARY_ADD', asm)
 
+    def test_constant_folding(self):
+        # Issue #11244: aggressive constant folding.
+        exprs = [
+            "3 * -5",
+            "-3 * 5",
+            "2 * (3 * 4)",
+            "(2 * 3) * 4",
+            "(-1, 2, 3)",
+            "(1, -2, 3)",
+            "(1, 2, -3)",
+            "(1, 2, -3) * 6",
+            "lambda x: x in {(3 * -5) + (-1 - 6), (1, -2, 3) * 2, None}",
+        ]
+        for e in exprs:
+            asm = dis_single(e)
+            self.assertNotIn('UNARY_', asm, e)
+            self.assertNotIn('BINARY_', asm, e)
+            self.assertNotIn('BUILD_', asm, e)
+
 
 def test_main(verbose=None):
     import sys

Misc/NEWS

@@ -10,6 +10,10 @@ What's New in Python 3.3 Alpha 1?
 Core and Builtins
 -----------------
 
+- Issue #11244: The peephole optimizer is now able to constant-fold
+  arbitrarily complex expressions.  This also fixes a 3.2 regression where
+  operations involving negative numbers were not constant-folded.
+
 - Issue #11450: Don't truncate hg version info in Py_GetBuildInfo() when
   there are many tags (e.g. when using mq).  Patch by Nadeem Vawda.