Comment (by Erik Schnetter):

I wonder what triggers the problem. Are there multiple concurrent iterators? Is the iterator used in multiple threads? Is the bboxset2 set changed while an iterator is running? Or do I misunderstand the definition of mutable?

Disabling bboxset2 is very expensive when running on many MPI processes. Since machines today have many more cores than they used to, I would be careful with that.

My guess is that any failure in a low-level mechanism such as bboxset2 will lead to instant death for any simulation.

--
Ticket URL: https://bitbucket.org/einsteintoolkit/tickets/issues/2426/g-10-possibly-miscompiles-parts-of-carpet