Re: [IMP-dev] Particle removal and deactivation

[Ben Webb <ben@salilab.org>]

Daniel Russel wrote:
> Ben Webb wrote:
> > The testcase tried to use a Python reference to a particle which had 
> > already been deleted. This clearly won't work, which is why I fixed it 
> > in the patch. Running that testcase through valgrind clearly showed 
> > access to freed memory, as I wrote in the svn log. By all means, 
> > submit a patch which fixes the bug if you like. But the reference 
> > counting clearly does not work with Python right now, and I don't have 
> > time to fix this the "right" way.
> No. The nbl should have been keeping the particle alive until the 
> evaluate call (and the evaluate call will have memory issues if the 
> memory is no longer valid since it needs to check if the particle is 
> active or not). So there are probably deeper bugs.
Indeed. You can reproduce the problem with the following:
% cd imp/kernel/test/states
% svn up -r580 test_nonbonded_list.py
% ../../../bin/imppy.sh valgrind 
--suppressions=../../../tools/valgrind-python.supp python 
test_nonbonded_list.py -v > /dev/null
With latest SVN, this will run through and report 0 errors. With r580, 
you'll get stuff like the following:
Test bbox NBL ... ==10105== Invalid read of size 1
==10105==    at 0x46C509B: _wrap_Particle_get_is_active (IMP_wrap.cc:3544)
==10105==    by 0x27D758C: PyCFunction_Call (in 
/usr/lib/libpython2.5.so.1.0)
==10105==    by 0x27A30B6: PyObject_Call (in /usr/lib/libpython2.5.so.1.0)
==10105==    by 0x2825E7E: PyEval_EvalFrameEx (in 
/usr/lib/libpython2.5.so.1.0)
==10105==    by 0x2828D74: PyEval_EvalCodeEx (in 
/usr/lib/libpython2.5.so.1.0)
==10105==    by 0x2827448: PyEval_EvalFrameEx (in 
/usr/lib/libpython2.5.so.1.0)
==10105==    by 0x282862E: PyEval_EvalFrameEx (in 
/usr/lib/libpython2.5.so.1.0)
==10105==    by 0x282862E: PyEval_EvalFrameEx (in 
/usr/lib/libpython2.5.so.1.0)
==10105==    by 0x282862E: PyEval_EvalFrameEx (in 
/usr/lib/libpython2.5.so.1.0)
==10105==    by 0x2828D74: PyEval_EvalCodeEx (in 
/usr/lib/libpython2.5.so.1.0)
==10105==    by 0x27C2EB1: (within /usr/lib/libpython2.5.so.1.0)
==10105==    by 0x27A30B6: PyObject_Call (in /usr/lib/libpython2.5.so.1.0)
==10105==  Address 0x4a162ac is 20 bytes inside a block of size 76 free'd
==10105==    at 0x40054AA: operator delete(void*) (vg_replace_malloc.c:342)
==10105==    by 0x48AC2B8: IMP::Particle::~Particle() (Particle.cpp:23)
==10105==    by 0x48A6573: IMP::internal::ObjectContainer<IMP::Particle, 
IMP::Index<IMP::ParticleTag> >::unref(IMP::Particle*) (ObjectContainer.h:55)
==10105==    by 0x48AAB6E: IMP::internal::ObjectContainer<IMP::Particle, 
IMP::Index<IMP::ParticleTag> >::remove(IMP::Index<IMP::ParticleTag>) 
(ObjectContainer.h:95)
==10105==    by 0x48A3666: 
IMP::Model::remove_particle(IMP::Index<IMP::ParticleTag>) (Model.cpp:34)
==10105==    by 0x46B8168: _wrap_Model_remove_particle (IMP_wrap.cc:14118)
==10105==    by 0x27D758C: PyCFunction_Call (in 
/usr/lib/libpython2.5.so.1.0)
==10105==    by 0x27A30B6: PyObject_Call (in /usr/lib/libpython2.5.so.1.0)
==10105==    by 0x2825E7E: PyEval_EvalFrameEx (in 
/usr/lib/libpython2.5.so.1.0)
==10105==    by 0x2828D74: PyEval_EvalCodeEx (in 
/usr/lib/libpython2.5.so.1.0)
==10105==    by 0x2827448: PyEval_EvalFrameEx (in 
/usr/lib/libpython2.5.so.1.0)
==10105==    by 0x282862E: PyEval_EvalFrameEx (in 
/usr/lib/libpython2.5.so.1.0)
==10105==

	Ben
--
ben@salilab.org                      http://salilab.org/~ben/
"It is a capital mistake to theorize before one has data."
	- Sir Arthur Conan Doyle