PyThreadState_Swap crash

[Philip Semanchuk]

On Apr 4, 2011, at 9:08 AM, Wiktor Adamski wrote:

> I have 2 threads in C code using python 2.5.2. First thread creates
> new interpreter (i need several interpreters but those 2 threads use
> only one) like that:
> 
> PyEval_AcquireLock();
> threadState = Py_NewInterpreter();
> PyThreadState_Swap(threadState);
> 
> // calling python API
> 
> PyThreadState_Swap(NULL);
> PyEval_ReleaseLock();
> 
> Second thread uses interpreter created in first thread:
> 
> PyEval_AcquireLock();
> PyThreadState_Swap(threadState);
> 
> and sometimes PyThreadState_Swap crashes in debug build
> (PyGILState_GetThisThreadState() returns garbage). In release build
> that code doesn't run and so far no other problem was found.
> I call PyEval_InitThreads() at the begining of program and every
> PyEval_AcquireLock() has PyEval_ReleaseLock().
> 
> Am I allowed to use the same threadState in different threads?
> If I am, is there another problem in my code?
> Or maybe it's a bug in python - acording to documentation "Python
> still supports the creation of additional interpreters (using
> Py_NewInterpreter()), but mixing multiple interpreters and the
> PyGILState_*() API is unsupported." - I don't use PyGILState_ but it's
> used internally in PyThreadState_Swap(). I also don't use
> PyEval_RestoreThread() - comment sugests that crashing code is present
> because possibility of calling from PyEval_RestoreThread().

Hi Wiktor,
I'm sorry I don't have a solution or even a suggestion for you. I just wanted to point out that PyEval_AcquireLock() and PyEval_ReleaseLock() were recently deprecated:
http://bugs.python.org/issue10913

Obviously they'll be around for quite a while longer but given the ominous-but-vague warning in issue10913's description, you might want to stay away from them. It's frustrating for me because I've got code I can't get to work without them.

Good luck
Philip