Core dump depending on simulation/lattice variable scope

[e-carlin]

Hi - I'm running into a core dump issue using synergia. It seems to depend on the scope of the simulation and/or lattice variables.

Steps to reproduce:

• compile from synergia3-2022.11

• apply the below patch to the wiki example. This patch is a simple contrived example to show the bug. In reality I have code generating a simulation and it generates a script with all code at the outer most scope of the script (no functions).

  
  ~$ patch synergia-example-from-web.py <<'EOF'
  > @@ -98,14 +98,15 @@
  
   # save
   synergia.simulation.checkpoint_save(propagator, sim)

• return sim

  def main():

  try:

• run()
• return run() except: raise RuntimeError("Failure to launch fodo.run")

-main() +sim = main()

EOF

- See error:

~$ python synergia-example-from-web.py Propagator: starting turn 1, final turn 10

Propagator: maximum number of turns reached Propagator: total time = 5.293s terminate called after throwing an instance of 'std::runtime_error' what(): Kokkos allocation "corr" is being deallocated after Kokkos::finalize was called ```

Turning on DEBUG level logging in the simlog object doesn't yield any additional information about the possible cause.

Do you have any idea what might be causing this? Is there a way to resolve it?

Let me know if I can provide more info. Thanks for your help.

[egstern]

I think what you are seeing is that all classes and objects that hold Kokkos objects have to be destroyed either by going out-of-scope or explicit del in Python before the program terminates. It looks like you are trying to sneak sim object out of the run() into the top level.

[e-carlin]

Aha, thanks @egstern. So it is about the scoping. I guess I'll modify the generation code to cope the variables or do a del. I appreciate the speedy reply.