[pypy-commit] pypy py3.5: officialize a CPython difference

[arigo]

Author: Armin Rigo <arigo at tunes.org>
Branch: py3.5
Changeset: r89577:b7305d7e2e07
Date: 2017-01-15 11:26 +0100
http://bitbucket.org/pypy/pypy/changeset/b7305d7e2e07/

Log:	officialize a CPython difference

diff --git a/lib-python/3/test/test_finalization.py b/lib-python/3/test/test_finalization.py
--- a/lib-python/3/test/test_finalization.py
+++ b/lib-python/3/test/test_finalization.py
@@ -171,7 +171,12 @@
             gc.collect()
             self.assert_del_calls(ids)
             self.assert_survivors(ids)
-            self.assertIsNot(wr(), None)
+            if support.check_impl_detail():
+                # CPython >= 3.4 changed this, but not in all cases
+                # (see the similar test in SelfCycleFinalizationTest).
+                # PyPy cannot tell apart the two cases, so we picked the
+                # cycle behavior, as it is the historical one.
+                self.assertIsNot(wr(), None)
             self.clear_survivors()
             gc.collect()
             self.assert_del_calls(ids)
diff --git a/pypy/doc/cpython_differences.rst b/pypy/doc/cpython_differences.rst
--- a/pypy/doc/cpython_differences.rst
+++ b/pypy/doc/cpython_differences.rst
@@ -166,6 +166,13 @@
 .. __: https://docs.python.org/2/library/weakref.html
 .. __: https://bitbucket.org/pypy/pypy/issue/2030/
 
+A new difference: before CPython 3.4, a weakref to ``x`` was always
+cleared before the ``x.__del__()`` method was called.  Since CPython 3.4
+the picture is more muddy.  Often, the weakref is still alive while
+``x.__del__()`` runs, but not always (e.g. not in case of reference
+cycles).  In PyPy3 we have kept the more consistent pre-3.4 behavior; we
+can't do something really different if there are cycles or not.
+
 ---------------------------------
 
 There are a few extra implications from the difference in the GC.  Most