segfault at exit when an exception exited stage_version context

[quarl]

$ cat repro.py
import datetime
import h5py
import numpy
from   versioned_hdf5           import VersionedHDF5File

import some_deshaw_module

def f():
    path_to_write = '/tmp/foo.h5'
    timestamp = numpy.datetime64(datetime.datetime.utcnow())
    f = h5py.File(path_to_write, 'w')
    vf = VersionedHDF5File(f)
    with vf.stage_version('v1', timestamp=timestamp) as sv:
        sv.create_dataset('foo', data=numpy.arange(10))
    with vf.stage_version('v2', timestamp=timestamp) as sv:
        1/0

f()

$ ipython repro.py
ZeroDivisionError: division by zero
zsh: segmentation fault (core dumped)

If I remove the 'import some_deshaw_module' or use python instead of ipython, then it doesn't reproduce the segfault.
I suspect whether the segfault happens depends on specifics of memory layout, so even if we get a reproducer without some_deshaw_module it might still not segfault on a different computer. I've seen similar cases where even the length of the contents of $PATH affected whether code segfaulted.

Before we do more work on making a more self-contained reproducer, are you able to see what the problem could be based on inspecting the code?

[quarl]

(internal ticket: https://desflow.nyc.deshaw.com/q/PyInf/10233)

[ArvidJB]

Here's the stack trace:

(gdb) bt
#0  0x00007ffff7e900e4 in ?? ()
#1  0x00007fffee5974dc in H5P_close () from /usr/local/python/python-3.10/std/lib/libhdf5.so.310
#2  0x00007fffee5978b9 in H5P__close_list_cb () from /usr/local/python/python-3.10/std/lib/libhdf5.so.310
#3  0x00007fffee51f7a2 in H5I_dec_ref () from /usr/local/python/python-3.10/std/lib/libhdf5.so.310
#4  0x00007fffee473ed2 in H5D__virtual_reset_layout () from /usr/local/python/python-3.10/std/lib/libhdf5.so.310
#5  0x00007fffee5605bd in H5O__layout_reset () from /usr/local/python/python-3.10/std/lib/libhdf5.so.310
#6  0x00007fffee563659 in H5O_msg_reset () from /usr/local/python/python-3.10/std/lib/libhdf5.so.310
#7  0x00007fffee572fd1 in H5P__dcrt_layout_close () from /usr/local/python/python-3.10/std/lib/libhdf5.so.310
#8  0x00007fffee597566 in H5P_close () from /usr/local/python/python-3.10/std/lib/libhdf5.so.310
#9  0x00007fffee5978b9 in H5P__close_list_cb () from /usr/local/python/python-3.10/std/lib/libhdf5.so.310
#10 0x00007fffee51d959 in H5I_clear_type () from /usr/local/python/python-3.10/std/lib/libhdf5.so.310
#11 0x00007fffee59435e in H5P_term_package () from /usr/local/python/python-3.10/std/lib/libhdf5.so.310
#12 0x00007fffee3e2dcc in H5_term_library.part.0 () from /usr/local/python/python-3.10/std/lib/libhdf5.so.310
#13 0x00007ffff69341ec in __run_exit_handlers () from /lib64/libc.so.6
#14 0x00007ffff6934320 in exit () from /lib64/libc.so.6
#15 0x00007ffff7a4157b in Py_Exit (sts=1) at /usr/src/debug/deshaw-python310-3.10.4-1.el8.3.deshaw.x86_64/Python/pylifecycle.c:2862
#16 0x00007ffff7a3e67b in handle_system_exit () at /usr/src/debug/deshaw-python310-3.10.4-1.el8.3.deshaw.x86_64/Python/pythonrun.c:775
#17 0x00007ffff7a3e5fd in _PyErr_PrintEx (set_sys_last_vars=1, tstate=0x620920) at /usr/src/debug/deshaw-python310-3.10.4-1.el8.3.deshaw.x86_64/Python/pythonrun.c:785
#18 PyErr_PrintEx (set_sys_last_vars=1) at /usr/src/debug/deshaw-python310-3.10.4-1.el8.3.deshaw.x86_64/Python/pythonrun.c:880
#19 0x00007ffff7a3665a in PyErr_Print () at /usr/src/debug/deshaw-python310-3.10.4-1.el8.3.deshaw.x86_64/Python/pythonrun.c:886
#20 _PyRun_SimpleFileObject (fp=<optimized out>, filename=<optimized out>, closeit=<optimized out>, flags=0x7fffffffd418)
    at /usr/src/debug/deshaw-python310-3.10.4-1.el8.3.deshaw.x86_64/Python/pythonrun.c:462
#21 0x00007ffff7a36187 in _PyRun_AnyFileObject (fp=0x7539a0, filename='/usr/local/bin/ipython', closeit=1, flags=0x7fffffffd418)
    at /usr/src/debug/deshaw-python310-3.10.4-1.el8.3.deshaw.x86_64/Python/pythonrun.c:90
#22 0x00007ffff7a3350d in pymain_run_file_obj (skip_source_first_line=0, filename='/usr/local/bin/ipython', program_name='/usr/local/python/python-3.10/std/bin/python')
    at /usr/src/debug/deshaw-python310-3.10.4-1.el8.3.deshaw.x86_64/Modules/main.c:366
#23 pymain_run_file (config=0x604d80) at /usr/src/debug/deshaw-python310-3.10.4-1.el8.3.deshaw.x86_64/Modules/main.c:385
#24 pymain_run_python (exitcode=0x7fffffffd410) at /usr/src/debug/deshaw-python310-3.10.4-1.el8.3.deshaw.x86_64/Modules/main.c:600
#25 Py_RunMain () at /usr/src/debug/deshaw-python310-3.10.4-1.el8.3.deshaw.x86_64/Modules/main.c:679
#26 0x00007ffff7a000cd in Py_BytesMain (argc=<optimized out>, argv=<optimized out>) at /usr/src/debug/deshaw-python310-3.10.4-1.el8.3.deshaw.x86_64/Modules/main.c:733
#27 0x00007ffff691dca3 in __libc_start_main () from /lib64/libc.so.6
#28 0x000000000040076e in _start ()

[peytondmurray]

I'm working on this now, but can't seem to reproduce the issue. Do you know of a substitute module for some_deshaw_module?

I'm also curious whether the timestamp kwarg is important. For now I'll try some other imports and see if I can't get this to segfault. So far I've only gotten ZeroDivisionError.