Module unloading resulting from SystemExit exception

[Patrick Hartling]

I am trying to understand how the SystemExit exception influences
module unloading. The situation that I am facing is that a Python
application behaves just fine upon exiting when executed using the
standard Python interpreter, but when run as a frozen executable,
modules are unloaded unexpectedly as a result of sys.exit() being
called. This wouldn't be an issue at all except that callbacks
registered with the atexit module have to re-import all modules that
they want to use, and this can be rather unwieldy.

I have narrowed it down to the SystemExit exception by determining
that the problem does not occur if my __main__ module catches
SystemExit and allows the script to terminate by reaching the end of
__main__. If the exception is not caught or is re-raised, then the
problem occurs. Since SystemExit influences the shutdown of threads,
it seems that I need to let it be raised to keep the application from
deadlocking on exit.

I am working with Python 2.6.1, and I saw this behavior with 2.5 as
well. It happens on Windows, Mac OS X, and Linux. The code that I am
using for the frozen executable is generated by sib.py from VendorID
1.0.0. The fact that VendorID is being used is inconsequential. I have
verified that its use in the executable is not causing this behavior
by removing all references to it. I have reduced this to the minimum
amount of code required to reproduce the problem and attached the
Python script an the C code generated by sib.py for the application
main() function.

My guess is that the code that sib.py generates is failing to do
something that Py_Main() would normally do. To that end, I have tried
modifying the main() function in a variety of ways to get it as close
to Py_Main() as I can, but it still exhibits the undesirable behavior
when sys.exit() is called. That pretty much leaves me thinking that
PyImport_ExecCodeModule() or PyImport_ImportFrozenModule() (I have
also looked at Py_FrozenMain() for insight) is the culprit.

Can anyone point me in the right direction for understanding how the
SystemExit exception results in this behavior? My hope is that there
is a change (simple or not) that can be made to the C code for the
frozen executable to resolve the problem.

Thanks in advance.

 -Patrick


-- 
Patrick L. Hartling
http://www.137.org/patrick/
-------------- next part --------------
/*==========================================================================*/
/* Main startup code for test                                               */
/* This is generated code; Do not modify it!                                */
/*--------------------------------------------------------------------------*/
#include <stdarg.h>
#include <Python.h>
#include <marshal.h>
/* #include <vendorid.h> */

extern unsigned char M___main__[];

static struct _frozen _PyImport_FrozenModules[] =
{
    {0, 0, 0} /* sentinel */
};

int main(int argc, char **argv)
{
    PyObject *v, *co;
    PyImport_FrozenModules = _PyImport_FrozenModules;

    Py_SetProgramName((char *)"test");
/*    vendorid_init(); */
    Py_Initialize();
    PySys_SetArgv(argc, argv);
    co = PyMarshal_ReadObjectFromString((char *)M___main__, 655);
    v = PyImport_ExecCodeModule((char *)"__main__", co);
    if(v == NULL)
    {
        PyErr_Print();
        return -1;
    }
    Py_DECREF(v);
    Py_Finalize();
    return 0;
}

/*==========================================================================*/
/* EOF                                                                      */
/*--------------------------------------------------------------------------*/
-------------- next part --------------
A non-text attachment was scrubbed...
Name: test.py
Type: text/x-python
Size: 243 bytes
Desc: not available
URL: <http://mail.python.org/pipermail/python-list/attachments/20090206/e89e5597/attachment.py>