Bug 794995

Summary: Memory leak in py_object_call found in lunatic-python-1.0.1-0.6.20090917bzr.fc17 using gcc-with-cpychecker static analyzer
Product: [Fedora] Fedora Reporter: Dave Malcolm <dmalcolm>
Component: lunatic-pythonAssignee: Gwyn Ciesla <limburgher>
Status: CLOSED RAWHIDE QA Contact: Fedora Extras Quality Assurance <extras-qa>
Severity: unspecified Docs Contact:
Priority: unspecified    
Version: rawhideCC: limburgher
Target Milestone: ---   
Target Release: ---   
Hardware: Unspecified   
OS: Unspecified   
URL: http://fedorapeople.org/~dmalcolm/gcc-python-plugin/2012-02-18/lunatic-python-1.0.1-0.6.20090917bzr.fc17/
Whiteboard:
Fixed In Version: Doc Type: Bug Fix
Doc Text:
Story Points: ---
Clone Of: Environment:
Last Closed: 2012-02-20 11:42:00 EST Type: ---
Regression: --- Mount Type: ---
Documentation: --- CRM:
Verified Versions: Category: ---
oVirt Team: --- RHEL 7.3 requirements from Atomic Host:
Cloudforms Team: ---
Bug Depends On:    
Bug Blocks: 789472    

Description Dave Malcolm 2012-02-18 10:30:03 EST
Description of problem:
I've been writing an experimental static analysis tool to detect bugs commonly occurring within C Python extension modules:
  https://fedorahosted.org/gcc-python-plugin/
  http://gcc-python-plugin.readthedocs.org/en/latest/cpychecker.html
  http://fedoraproject.org/wiki/Features/StaticAnalysisOfPythonRefcounts

I ran the latest version of the tool (in git master; post 0.9) on
lunatic-python-1.0.1-0.6.20090917bzr.fc17.src.rpm, and it reports various errors.

You can see a list of errors here, triaged into categories (from most significant to least significant):
http://fedorapeople.org/~dmalcolm/gcc-python-plugin/2012-02-18/lunatic-python-1.0.1-0.6.20090917bzr.fc17/

I've manually reviewed the issues reported by the tool.

Within the category "Reference leaks" the 1 issues reported
  src/pythoninlua.c:py_object_call:ob_refcnt of '*args' is 1 too high
appears to be a genuine bug: the "args" is tuple is allocated, populated, and passed to PyObject_CallObject(), but nothing seems to do a Py_DECREF() on it, hence the tuple of arguments becomes immortal for the rest of the lifetime of the process.  If I'm reading this correctly, every call to py_object_call will leak memory.

There may of course be other bugs in my checker tool.

Hope this is helpful; let me know if you need help reading the logs that the tool generates - I know that it could use some improvement.

Version-Release number of selected component (if applicable):
lunatic-python-1.0.1-0.6.20090917bzr.fc17
gcc-python-plugin post-0.9 git 771455b3128b1323e80bdda53939d8d140a84c68 running the checker in an *f16* chroot
Comment 1 Gwyn Ciesla 2012-02-20 10:17:38 EST
It looks like inserting the following at line 138 would do the trick:

for (i = 0; i != nargs; i++) {
  Py_DECREF(args);
}

Would it not?
Comment 2 Dave Malcolm 2012-02-20 11:22:44 EST
No, it just needs it once, I think, rather than nargs times: "args" is the whole tuple, which the code owns one reference to.

I think this code just needs a single extra Py_DECREF(args) at around line 137, like this:

     value = PyObject_CallObject(obj->o, args);
+    Py_DECREF(args);
     if (value) {
Comment 3 Gwyn Ciesla 2012-02-20 11:41:46 EST
Got it, thanks!  Fixed in rawhide.
Comment 4 Dave Malcolm 2012-02-20 11:55:48 EST
Thanks!  Can you make sure upstream get this report also?
Comment 5 Gwyn Ciesla 2012-02-20 12:10:10 EST
Will do.