10:27 a.m.
This changes CPython.change_refcount to also look for a WithinRange
object. This is needed because UnknownValue.make may return a
WithinRange object in some cases.
This fixes issue #46.
---
libcpychecker/refcounts.py | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/libcpychecker/refcounts.py b/libcpychecker/refcounts.py
index abe034a..8d2228b 100644
--- a/libcpychecker/refcounts.py
+++ b/libcpychecker/refcounts.py
@@ -549,7 +549,7 @@ class CPython(Facet):
check_isinstance(oldvalue, AbstractValue)
log('oldvalue: %r', oldvalue)
# If we never had a ob_refcnt, treat it as a borrowed reference:
- if isinstance(oldvalue, UnknownValue):
+ if isinstance(oldvalue, UnknownValue) or isinstance(oldvalue, WithinRange):
oldvalue = RefcountValue.borrowed_ref(loc, pyobjectptr.region)
check_isinstance(oldvalue, RefcountValue)
newvalue = fn(oldvalue)
--
1.8.1.4
10:45 a.m.
>>>> "Tom" == Tom Tromey
<tromey(a)redhat.com> writes:
Tom> This changes CPython.change_refcount to also look for a WithinRange
Tom> object. This is needed because UnknownValue.make may return a
Tom> WithinRange object in some cases.
Tom> This fixes issue #46.
I spoke too soon. Not sure what I was thinking there...
I think the test case is correct but with this patch the checker still
emits an error. So, while it eliminates the unhandled exception, there
is still a bug somewhere.
Tom
11:51 a.m.
Here is try #2.
Tom
From 2de0ba029f3b5d7ec2353a13385e648b3ace54ec Mon Sep 17 00:00:00 2001
From: Tom Tromey <tromey(a)redhat.com>
Date: Thu, 18 Apr 2013 09:25:59 -0600
Subject: [PATCH] fix issue #46
This fixes issue #46.
The bug is that State.var_region checks for PyObject globals, but it
should also check for derived types like PyTypeObject. The fix is to
use the existing type_is_pyobjptr_subclass function to do the check.
This still gives an error for the test case in the issue -- but that
test case leaks a reference in the situation where PyModule_AddObject
fails.
---
libcpychecker/absinterp.py | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/libcpychecker/absinterp.py b/libcpychecker/absinterp.py
index c66e9ed..d11de19 100644
--- a/libcpychecker/absinterp.py
+++ b/libcpychecker/absinterp.py
@@ -1634,7 +1634,8 @@ class State(object):
# Initialize the refcount of global PyObject instances
# e.g. _Py_NoneStruct to 0 i.e. we don't own any references to them
- if str(var.type) == 'struct PyObject':
+ from libcpychecker.refcounts import type_is_pyobjptr_subclass
+ if type_is_pyobjptr_subclass(var.type.pointer):
from libcpychecker.refcounts import RefcountValue
ob_refcnt = self.make_field_region(region, 'ob_refcnt') # FIXME:
this should be a memref and fieldref
self.value_for_region[ob_refcnt] = RefcountValue.borrowed_ref(None,
region)
--
1.8.1.4
12:20 p.m.
>>>> "Tom" == Tom Tromey
<tromey(a)redhat.com> writes:
Tom> Here is try #2.
... and of course it is wrong too.
I forgot to run the test suite for this one, which would have shown that
it is no good. Sorry about that.
With this patch in place, the checker starts complaining about incorrect
refcounts for types, e.g.:
../../archer/gdb/python/python.c: In function ‘gdbpy_parameter_value’:
../../archer/gdb/python/python.c:458:1: error: ob_refcnt of PyLong_Type is 1 too low
[-Werror]
../../archer/gdb/python/python.c:458:1: note: was expecting final ob_refcnt to be N + 1
(for some unknown N)
../../archer/gdb/python/python.c:458:1: note: due to object being referenced by:
PyLongObject.ob_type
../../archer/gdb/python/python.c:458:1: note: but final ob_refcnt is N + 0
../../archer/gdb/python/python.c:404:3: note: when following case 3 at: switch (type)
../../archer/gdb/python/python.c:440:12: note: when treating unknown void * from
../../archer/gdb/python/python.c:402 as non-NULL at: if ((* (int *) var) ==
INT_MAX)
../../archer/gdb/python/python.c:440:10: note: when considering range: -0x80000000 <=
value <= 0x7ffffffe at: if ((* (int *) var) == INT_MAX)
../../archer/gdb/python/python.c:440:10: note: taking False path at: if ((* (int *)
var) == INT_MAX)
../../archer/gdb/python/python.c:444:30: note: ob_refcnt is now refs: 0 + N where N >=
1
../../archer/gdb/python/python.c:444:7: note: when PyLong_FromLong() succeeds at:
return PyLong_FromLong (* (int *) var);
../../archer/gdb/python/python.c:444:7: note: PyLong_Type is now referenced by 1 non-stack
value(s): PyLongObject.ob_type
So something is wrong with the modeling of ob_type.
I don't know how to fix that.
Tom
4050
days inactive
4050
days old
gcc-python-plugin@lists.fedorahosted.org
3 comments
1 participants
participants (1)
-
Tom Tromey