[Python-checkins] r69067 - in python/branches/release30-maint/Lib/test/crashers: loosing_mro_ref.py losing_mro_ref.py

raymond.hettinger python-checkins at python.org
Wed Jan 28 22:06:01 CET 2009


Author: raymond.hettinger
Date: Wed Jan 28 22:06:01 2009
New Revision: 69067

Log:
Fix spelling in filename.

Added:
   python/branches/release30-maint/Lib/test/crashers/losing_mro_ref.py
      - copied unchanged from r69066, /python/branches/release30-maint/Lib/test/crashers/loosing_mro_ref.py
Removed:
   python/branches/release30-maint/Lib/test/crashers/loosing_mro_ref.py

Deleted: python/branches/release30-maint/Lib/test/crashers/loosing_mro_ref.py
==============================================================================
--- python/branches/release30-maint/Lib/test/crashers/loosing_mro_ref.py	Wed Jan 28 22:06:01 2009
+++ (empty file)
@@ -1,35 +0,0 @@
-"""
-There is a way to put keys of any type in a type's dictionary.
-I think this allows various kinds of crashes, but so far I have only
-found a convoluted attack of _PyType_Lookup(), which uses the mro of the
-type without holding a strong reference to it.  Probably works with
-super.__getattribute__() too, which uses the same kind of code.
-"""
-
-class MyKey(object):
-    def __hash__(self):
-        return hash('mykey')
-
-    def __eq__(self, other):
-        # the following line decrefs the previous X.__mro__
-        X.__bases__ = (Base2,)
-        # trash all tuples of length 3, to make sure that the items of
-        # the previous X.__mro__ are really garbage
-        z = []
-        for i in range(1000):
-            z.append((i, None, None))
-        return 0
-
-
-class Base(object):
-    mykey = 'from Base'
-
-class Base2(object):
-    mykey = 'from Base2'
-
-# you can't add a non-string key to X.__dict__, but it can be
-# there from the beginning :-)
-X = type('X', (Base,), {MyKey(): 5})
-
-print(X.mykey)
-# I get a segfault, or a slightly wrong assertion error in a debug build.


More information about the Python-checkins mailing list