Could you post the full traceback of the original crash? Set only BOOST_ADAPTBX_FPE_DEFAULT but not BOOST_ADAPTBX_SIGNALS_DEFAULT. That should tell you where the segfault originates. Another approach is to compile everything with -g and use valgrind. That will even give you line numbers. However, it may be tricky to make this work. Hopefully the traceback will give enough of a clue. Ralf ________________________________ From: Cameron Mura To: Ralf W. Grosse-Kunstleve ; [email protected]; Pymol-users Sent: Sunday, July 5, 2009 9:14:23 PM Subject: Re: [cctbxbb] draw_symops <-- cctbx(/phenix) broken due to boost segfault handling ? Hi Ralf, Thanks for your note...

I tried the recommendation of setting the BOOST_ADAPTBX* env vars, but this doesn't remedy the problem.

...

Did you get the same traceback? -- That would mean the env vars aren't defined in the right context. With the env vars defined, there shouldn't be any tracebacks anymore, just plain "Abort". Right -- This is what happened. I verified that the pymol context was aware of these env vars:

PyMOL>import os

...

PyMOL>print os.environ.get('BOOST_ADAPTBX_FPE_DEFAULT') 1 PyMOL>print os.environ.get('BOOST_ADAPTBX_SIGNALS_DEFAULT') 1

So... with the vars set, no backtrace... just a simple crash: <snip>....PyMOL

session....</snip>

...

...... COMPND 6 ENGINEERED: YES; COMPND 7 MUTATION: YES ...... Symmetry: Found 12 symmetry operators. CmdLoad: "1avv.pdb" loaded as "1avv". PyMOL>run draw_symops_cctbx.py Finished importing for draw_symops_cctbx.py PyMOL>draw_symops 1avv /usr/bin/pymol: line 4: 4756 Segmentation fault cctbx.python /usr/lib64/python2.5/site-packages/pymol/__init__.py $* $]

You snipped out

the most interesting part of the traceback. The first part is always the Python startup right, the red part below(?) [up to frame 1/19]

and anything

after boost::python::handle_exception is just the code producing the tracebacks. right, blue part below(?) [beyond libc's frame 13/19]

The important

clues are somewhere in the middle.

Following is what happens in a simpler test run, outside of pymol, just divide_by_zero.py from the command line (result is essentially identical to your Sep-08 post to python-dev): $] cctbx.python

/root/software/cctbx/2009_02_15_2320/working/cctbx_sources/boost_adaptbx/command_line/divide_by_zero.py

...

Now dividing by zero (in C++) ... show_stack(1): /root/software/cctbx/2009_02_15_2320/working/cctbx_sources/boost_adaptbx/command_line/divide_by_zero.py(11) run show_stack(2): /root/software/cctbx/2009_02_15_2320/working/cctbx_sources/boost_adaptbx/command_line/divide_by_zero.py(26) <module> libc backtrace (19 frames, most recent call last): /usr/bin/python [0x400649] /lib64/libc.so.6(__libc_start_main+0xe6) [0x344081e576] /usr/lib64/libpython2.5.so.1.0(Py_Main+0xb11) [0x34584e7ef1] /usr/lib64/libpython2.5.so.1.0(PyRun_SimpleFileExFlags+0x1cd) [0x34584ddfed] /usr/lib64/libpython2.5.so.1.0(PyRun_FileExFlags+0x96) [0x34584dca06] /usr/lib64/libpython2.5.so.1.0 [0x34584dc931] /usr/lib64/libpython2.5.so.1.0(PyEval_EvalCode+0x32) [0x34584c0aa2] /usr/lib64/libpython2.5.so.1.0(PyEval_EvalCodeEx+0x715) [0x34584c0865] /usr/lib64/libpython2.5.so.1.0(PyEval_EvalFrameEx+0x658d) [0x34584bfe6d] /usr/lib64/libpython2.5.so.1.0(PyEval_EvalFrameEx+0x36f3) [0x34584bcfd3] /usr/lib64/libpython2.5.so.1.0(PyObject_Call+0x13) [0x345843d493]

/root/software/cctbx/2009_02_15_2320/working/cctbx_build/lib/libboost_python.so [0x7f20bb3ba3b2]

...

/root/software/cctbx/2009_02_15_2320/working/cctbx_build/lib/libboost_python.so(boost::python::handle_exception_impl(boost::function0<void>)+0x85) [0x7f20bb3c6625]

/root/software/cctbx/2009_02_15_2320/working/cctbx_build/lib/libboost_python.so(boost::function0<void>::operator()() const+0x34) [0x7f20bb3c6f94]

...

/root/software/cctbx/2009_02_15_2320/working/cctbx_build/lib/libboost_python.so [0x7f20bb3bd398]

...

/root/software/cctbx/2009_02_15_2320/working/cctbx_build/lib/libboost_python.so(boost::python::objects::function::call(_object*, _object*) const+0x125) [0x7f20bb3bcec5]

...

/root/software/cctbx/2009_02_15_2320/working/cctbx_build/lib/boost_python_meta_ext.so(boost::python::objects::caller_py_function_impl >

...

::operator()(_object*, _object*)+0xf3) [0x7f20bb5f5333]

/root/software/cctbx/2009_02_15_2320/working/cctbx_build/lib/boost_python_meta_ext.so [0x7f20bb5f0a54]

...

/lib64/libc.so.6 [0x3440832f90] Floating-point error (Python and libc call stacks above) This crash may be due to a problem in any imported Python module, including modules which are not part of the cctbx project. To disable the traps leading to this message, define these environment variables (e.g. assign the value 1): BOOST_ADAPTBX_FPE_DEFAULT BOOST_ADAPTBX_SIGNALS_DEFAULT This will NOT solve the problem, just mask it, but may allow you to proceed in case it is not critical. $]

I'm still at somewhat of a loss (mainly due to ignorance about boost/python interactions), but will let you know if I solve it. thanks, cam === Ralf W. Grosse-Kunstleve wrote (on 07/05/2009 11:38 PM): === Hi Cameron,

...

There are quite a few 3rd party extensions causing floating-point exceptions (e.g. we have to turn off the floating-point traps when using wxPython).

...

...

I tried the recommendation of setting the BOOST_ADAPTBX* env vars, but this doesn't remedy the problem.

...

Did you get the same traceback? -- That would mean the env vars aren't defined in the right context. With the env vars defined, there shouldn't be any tracebacks anymore, just plain "Abort".

...

You snipped out the most interesting part of the traceback. The first part is always the Python startup, and anything after boost::python::handle_exception is just the code producing the tracebacks. The important clues are somewhere in the middle.

...

Ralf

________________________________ From: >Cameron Mura

To: [email protected]; Pymol-users Sent: Sunday, July 5, 2009 12:33:53 PM Subject: [cctbxbb] draw_symops <-- cctbx(/phenix) broken due to boost segfault handling ?

...

Hi all,

...

Apologies for the message to both mailing lists, but perhaps someone on either cctbxbb or pymol-users has had some recent experience with this...

...

The subject line more-or-less says it all: I'm wondering if anyone has recently used Rob Campbell's draw_symops (or draw_symops_cctbx) Python modules with a relatively recent (within past year) version of the cctbx (particularly the Boost C++ parts) ? I succeeded in doing this a few years ago to generate images such as the ones near the top of http://muralab.org/~cmura/PyMOL/Sandbox/, but I can't get it to work on my present system (64-bit Fedora 10; Python 2.5.2 with Python 2.6 coexisting).

...

Below is what happens when I...

...

1) Initiate a new PyMOL session (using a PyMOL v1.1 that I built in-place from source, so should be compatible with local python, system libs, etc)

...

2) load PDB files (tried with a few diff't files; verified that they contain proper CRYST specifications)

...

3) do the usual 'run draw_symops_cctbx.py'

...

4) call draw_symops() -- either implicitly, or by explicit specification of the u.c. params (i.e., using something like "draw_symops_param((45.2,45.2,70.8,90.,90.,120.),'p3121',0.5,0.1)")

...

....which crashes the pymol session and yields the following trace:

...

...

...

libc backtrace (48 frames, most recent call last): /usr/bin/python [0x400649] /lib64/libc.so.6(__libc_start_main+0xe6) [0x344081e576] /usr/lib64/libpython2.5.so.1.0(Py_Main+0xb11) [0x34584e7ef1] <snip> .... .... </snip> /usr/lib64/libpython2.5.so.1.0(PyObject_Call+0x13) [0x345843d493] /root/cctbx/working/v4/cctbx_build/lib/libboost_python.so [0x7ff19691db16]

/root/cctbx/working/v4/cctbx_build/lib/libboost_python.so(boost::python::handle_exception_impl(boost::function0<void>)+0x9f) [0x7ff19692a2ef]

...

/root/cctbx/working/v4/cctbx_build/lib/libboost_python.so [0x7ff196920b08]

/root/cctbx/working/v4/cctbx_build/lib/libboost_python.so(boost::python::objects::function::call(_object*, _object*) const+0x125) [0x7ff196920635]

...

/root/cctbx/working/v4/cctbx_build/lib/cctbx_sgtbx_ext.so(boost::python::detail::caller_arity<2u>::impl >::operator()(_object*, _object*)+0x12a) [0x7ff18f820a3a]

...

/root/cctbx/working/v4/cctbx_build/lib/cctbx_sgtbx_ext.so [0x7ff18f8034cc]

/root/cctbx/working/v4/cctbx_build/lib/libboost_python.so(boost::python::throw_error_already_set()+0xe) [0x7ff196929dde]

...

/usr/lib64/libstdc++.so.6(__cxa_allocate_exception+0x2f) [0x3444cc2cdf] /lib64/libc.so.6 [0x3440832f90] Segmentation fault (libc call stack above) This crash may be due to a problem in any imported Python module, including modules which are not part of the cctbx project. To disable the traps leading to this message, define these environment variables (e.g. assign the value 1): BOOST_ADAPTBX_FPE_DEFAULT BOOST_ADAPTBX_SIGNALS_DEFAULT This will NOT solve the problem, just mask it, but may allow you to proceed in case it is not critical. PyMOL: abrupt program termination. I tried the recommendation of setting the BOOST_ADAPTBX* env vars, but

PyMOL>draw_symops_param((45.2,45.2,70.8,90.,90.,120.),'p3121',0.5,0.1) this doesn't remedy the problem.

...

Also, I tried accomplishing this with many combinations of workflows: 1) old cctbx (e.g., 2005_04_29_1615), latest stable release of cctbx (2009_02_15_2320), bleeding-edge cctbx (2009_07_04_0143) 2) pre-built cctbx binaries versus building myself from src (for each of the above), insuring, again, optimal match between local system libs... 3) cctbx alone, cctbx in the context of Phenix, cctbx +/- its own bundled Python (2.5 and 2.6)

...

Also, I tried this with different combinations of Rob's scripts -- e.g., the 'symop_axes.dat'-dependent "draw_symops.py", versus the 'all_axes_new.py'-dependent "draw_symops_cctbx.py". Interestingly, I found that get_all_axes() from 'all_axes_new.py' works fine for different PDB files.

...

Investigating the above draw_symops stacktrace led me to this post from Ralf Grosse-Kunstleve to python-dev: http://mail.python.org/pipermail/python-dev/2008-September/082639.html It looks like exactly the same problem that's causing the pymol/draw_symops scripts to choke. Also, I can verify that issuing commands such as in Ralf's post (boost_adaptbx.segmentation_fault, boost_adaptbx.divide_by_zero) result in the same segfaults and errors as in his post, so... I believe this is the crux of it. Before delving into Boost codebase, cctbx's exception-handling in C++/Python Boost, etc., I'm wondering if anyone else has seen anything like this...? or has advice on how best to proceed? Any tips or suggestions would be most greatly appreciated.

...

Thanks, Cam