We are currently working on adding tracing support to Anaconda, the Fedora/Red Hat Enterprise Linux installer and we have encountered some pretty strange behavior of the trace module. Anaconda (or to be concrete the Blivet storage library used by Anaconda) uses the pyblock module, but just importing it crashes the trace module with a strange exception.

It can be reproduced like this:
0. install pyblock (on Fedora is is provided by the python-pyblock package)

1. write some python code that imports the "block" module provided by pyblock

echo "import block" > pyblock_trace.py

2. try to trace the code

python -m trace -t pyblock_trace.py

The trace module starts tracing but after a few seconds it crashes with the following traceback:


Traceback (most recent call last):
  File "/usr/lib64/python2.7/runpy.py", line 162, in _run_module_as_main
    "__main__", fname, loader, pkg_name)
  File "/usr/lib64/python2.7/runpy.py", line 72, in _run_code
    exec code in run_globals
  File "/usr/lib64/python2.7/trace.py", line 830, in <module>
    main()
  File "/usr/lib64/python2.7/trace.py", line 818, in main
    t.runctx(code, globs, globs)
  File "/usr/lib64/python2.7/trace.py", line 513, in runctx
    exec cmd in globals, locals
  File "pyblock_trace.py", line 1, in <module>
    import block
  File "/usr/lib64/python2.7/site-packages/block/__init__.py", line 47, in <module>
    import dmraid
  File "<string>", line 1, in <module>
  File "/usr/lib64/python2.7/trace.py", line 609, in globaltrace_lt
    filename = frame.f_globals.get('__file__', None)
AttributeError: 'module' object has no attribute 'get'


The dmraid module is written in C and we have looked though its source[1] code but have found nothing extraordinary. Most importantly there is no code touching the globals, but it still fails.

When looking what actually is in frame.f_globals we found that in all successful calls it has the globals dictionary but for the dmraid module it for some reason contains the module instance instead. Module instance is not a dictionary, so it doesn't have the get method and this leads to the exception above.

This is not the only C module we use, but this is the only one that triggers the crash in trace.


Additional information
Python version: 2.7.5
architecture: X86_64
OS: Fedora 20


[1] https://git.fedorahosted.org/cgit/pyblock.git/tree/dmraid.c

Third party compiled C crashers are nasty. Two suggestions:
1. Reduce the failing example to the minimum needed to fail. If you bypass all the python code in maps.py and device.py and just 'import dmraid', do you still get a crash at that point.
2. What happens if you run a minimal crasher with 3.4 or .5? I don't know if all the 3.x work on trace was backported.

@ 1.: Reproducer attached to the comment - just build the C code and run trace_test.py It is maybe not as minimal as it could be but I'm afraid the context of what the module is doing would be lost if it was cut down too aggressively.

@ 2.: I'm afraid this is not applicable in this case - pyblock just flat out does not support Python 3 and I haven't been able to find out even any third-party Python 3 port of it. I even tried to force run the current code with Python 3, just in case, but it just tracebacks during import due to Python 3 incompatible code, even before even importing the C extensions.

pyblock supports python 3 now. Are you seeing the problem pm 3.8+?Otherwise this issue can be closed since python 2 past EOL.

Martin, please create a new issue if this is still a problem in Python 3.