Segfault if starting a python algorithm while a script is running

[Janik Zikovsky]

Say you start a very simple script that takes a while:

for i in range(100):
    Pause(1)

and then start a python algorithm using the GUI. For example, I used ConjoinSpectra. You get a segfault with the following stack trace:

Thread [17] (Suspended: Signal 'SIGSEGV' received. Description: Segmentation fault.)	
	36 PyObject_Call() abstract.c:2530 0x00007ffff2984a5a	
	35 PyObject_CallFunctionObjArgs() abstract.c:2760 0x00007ffff2985410	
	34 PyObject_IsSubclass() abstract.c:3041 0x00007ffff2985c9d	
	33 PyErr_GivenExceptionMatches() errors.c:117 0x00007ffff2a40f8a	
	32 builtin_hasattr() bltinmodule.c:887 0x00007ffff2a27054	
	31 call_function() ceval.c:4013 0x00007ffff2a30630	
	30 PyEval_EvalFrameEx() ceval.c:2666 0x00007ffff2a30630	
	29 PyEval_EvalCodeEx() ceval.c:3253 0x00007ffff2a3227d	
	28 fast_function() ceval.c:4109 0x00007ffff2a3028d	
	27 call_function() ceval.c:4034 0x00007ffff2a3028d	
	26 PyEval_EvalFrameEx() ceval.c:2666 0x00007ffff2a3028d	
	25 PyEval_EvalCodeEx() ceval.c:3253 0x00007ffff2a3227d	
	24 fast_function() ceval.c:4109 0x00007ffff2a3028d	
	23 call_function() ceval.c:4034 0x00007ffff2a3028d	
	22 PyEval_EvalFrameEx() ceval.c:2666 0x00007ffff2a3028d	
	21 PyEval_EvalCodeEx() ceval.c:3253 0x00007ffff2a3227d	
	20 function_call() funcobject.c:526 0x00007ffff29b1192	
	19 PyObject_Call() abstract.c:2529 0x00007ffff2984a53	
	18 instancemethod_call() classobject.c:2578 0x00007ffff29978ff	
	17 PyObject_Call() abstract.c:2529 0x00007ffff2984a53	
	16 PyEval_CallObjectWithKeywords() ceval.c:3882 0x00007ffff2a2a627	
	15 PyEval_CallMethod() modsupport.c:586 0x00007ffff2a507de	
	14 boost::python::call_method<Mantid::PythonAPI::PyAlgorithmBase*>() call_method.hpp:66 0x00007fffb8091712	
	13 Mantid::PythonAPI::PyCall_NoArg<Mantid::PythonAPI::PyAlgorithmBase*>::dispatch() PythonInterfaceFunctions.h:133 0x00007fffb80914cb	
	12 Mantid::PythonAPI::PyAlgorithmCallback::clone() PyAlgorithmWrapper.cpp:50 0x00007fffb80900fa	
	11 Mantid::API::AlgorithmFactoryImpl::createAlgorithm() AlgorithmFactory.cpp:409 0x00007ffff6c59b19	
	10 Mantid::API::AlgorithmFactoryImpl::create() AlgorithmFactory.cpp:87 0x00007ffff6c580b8	
	9 Mantid::API::AlgorithmManagerImpl::createUnmanaged() AlgorithmManager.cpp:42 0x00007ffff6c657af	
	8 Mantid::API::AlgorithmProxy::createConcreteAlg() AlgorithmProxy.cpp:177 0x00007ffff6c6bf5e	
	7 Mantid::API::AlgorithmProxy::executeAsyncImpl() AlgorithmProxy.cpp:212 0x00007ffff6c6c19e	
	6 Poco::ActiveRunnable<bool, Poco::Void, Mantid::API::AlgorithmProxy>::run() ActiveRunnable.h:85 0x00007ffff6c6e6a0	
	5 Poco::PooledThread::run() ThreadPool.cpp:203 0x00007fffee79e185	
	4 Poco::ThreadImpl::runnableEntry() Thread_POSIX.cpp:345 0x00007fffee79acbd	
	3 start_thread() pthread_create.c:304 0x00007fffecb13d8c	
	2 clone() clone.S:112 0x00007fffec85f04d	
	1 <symbol is not available> 0x0000000000000000	

Perhaps we need some sort of lock to exclude two python "things" running at once? This will come up in LiveDataMonitoring, where we will have one thread running a python script for, say, a few seconds every few minutes. We want to be able to run a python script when the live data script is NOT running; conversely, we need LiveData to wait until any script started by the user finishes.

[Janik Zikovsky]

Note that I got a segfault with both the old and new python API's.

[Martyn Gigg]

It should be possible to run two bits of python simultaneously, as long as the appropriate Python locks are in place, it does the switching itself. It seems notoriously difficult to get right inside MantidPlot for some reason. It has definitely worked in the past so something has changed. I'll take a look.

[Martyn Gigg]

Fixed as part of #5006

[Roman Tolchenov]

Ran the script and ConjoinSpectra at the same time. Works fine.

[Stuart Campbell]

This ticket has been transferred to github issue ​5760