calldll and threads...

[Steve]

I could benefit from words of wisdom... this is the situation:

I want to use Python as a high level language to write a test suite
for a complex shared library (I've a windows DLL acting as proxy to
remote server(s)) which has several blocking calls, and for which the
interleaving and overlapping of calls is very significant. I need to
drive a client to demonstrate code coverage on the server side - there
is a large degree of server side interaction between concurrently
executing client actions.

In Python, my first attempt used thread.start_new_thread() when
executing a python wrapper for a call which could block.. which I
later rewrote to use threading.Thread.  Initially all seemed well
until I found that the whole python program blocked waiting for the
first blocking call to the DLL.  Then I read about the "Global
Interpreter Lock" and went on to look at the source for calldll.pyd -
which didn't seem to address threading.  I tried following the advice
in the python docs:

Py_BEGIN_ALLOW_THREADS
//...Do some blocking I/O operation... making sure not to use
return...
Py_END_ALLOW_THREADS

Which I used to wrapper all calls to dll functions with a customised
calldll... the above code massaged into call_foreign_function()... 
but I now find that while the very first call to my modified calldll
seems to work fine - subsequent calls crash the python interpreter.

Has anyone else had similar problems with calldll or anything like it?
Can anyone explain what I've done wrong with Py_BEGIN_ALLOW_THREADS
and friend? Do other people use python to script similar regression
test scenarios? Is there a standard mechanism to make a blocking call
from a python thread?

Thanks in advance,
Steve
[P.S. In case it's relevant. I'm using the Python 2.2 release from
www.python.org with the calldll source from www.nightmare.com running
under windows NT 4 with the Win32all build 146 using PythonWin as a
development environment]