Hanging in CI related to DAGMC test

[paulromano]

Bug Description

Recently we've been seeing a lot of CI jobs that hang right after running some DAGMC unit tests. Here's an example from a recent job:

tests/unit_tests/cell_instances/test_rect_multilattice.py::test_cell_instance_multilattice_results PASSED [  1%]
tests/unit_tests/dagmc/test.py::test_dagmc_temperatures[1-320.0] PASSED  [  1%]
tests/unit_tests/dagmc/test.py::test_dagmc_temperatures[2-300.0] PASSED  [  1%]
tests/unit_tests/dagmc/test.py::test_dagmc_temperatures[3-293.6] PASSED  [  2%]
tests/unit_tests/dagmc/test_bounds.py::test_bounding_box PASSED          [  2%]
tests/unit_tests/dagmc/test_bounds.py::test_bounding_region PASSED       [  2%]
tests/unit_tests/dagmc/test_bounds.py::test_bounded_universe PASSED      [  2%]
tests/unit_tests/dagmc/test_bounds.py::test_material_names PASSED        [  2%]
tests/unit_tests/dagmc/test_convert_to_multigroup.py::test_convert_to_multigroup_without_particles_batches PASSED [  2%]
tests/unit_tests/dagmc/test_h5m_subdir.py::test_model_h5m_in_subdirectory[True] PASSED [  2%]
tests/unit_tests/dagmc/test_h5m_subdir.py::test_model_h5m_in_subdirectory[False] PASSED [  2%]
<-- hangs here and eventually times out at 6 hours

I haven't been able to figure out why it hangs or reproduce it locally thus far. @pshriwise do you have any ideas on this?

[pshriwise]

Yeah I noted this as of #3801 being merged. I'll dig into it.

[paulromano]

To be clear, that hang has been happening sporadically for quite a while. Perhaps #3801 increased the likelihood of it?

[vitmog]

My #3816 does not contain #3801, but CI get stuck right now on the same stage.
UPD. No PR with failed CI contains #3801. I understand now.

[paulromano]

I asked both Claude and Codex to figure out what was going on with the hang and they both pointed to the same thing:

• In nested DAGMC mode, DAGCell::distance() does not mark a particle lost when no intersection is found in the implicit complement (unless the DAGMC universe is the root).
• If a particle escapes into void through the broken mesh, collision distance can become infinite, and it can keep streaming/reflecting.

I dropped a potential fix in #3817 which simply limits max_particle_events to avoid an infinite loop and that seems to have worked, but I might have also just gotten lucky. Let's keep an eye on CI and see if this shows up again. If not, we can close this issue.

[pshriwise]

Thanks @paulromano!

[GuySten]

Unfortunately, the CI still hangs...

https://github.com/openmc-dev/openmc/actions/runs/22404563088/job/64861461000?pr=3835

[paulromano]

That's disappointing 😞 back to the drawing board...