Segfault when <floors>/on = false and <fofc>/on = true

[phyzan]

I am testing a setup that avoids loading the floors package (floors/on = false) and also fofc/on = true.
However I get a segmentation fault at the very first step.

I think the issue is that even though the floors package is not loaded at all, the function KHARMADriver::AddFOFC in
kharma/driver/kharma_driver.cpp
calls

auto t_mark_floors = tl.AddTask(t_guess_prims, Floors::DetermineGRMHDFloors, guess, IndexDomain::entire, fofc_floors, fofc_floors_inner);

calling the floors package without performing a check.

Maybe the floors package should be loaded anyway, and something like this should be implemented in kharma/kharma.cpp?

    bool floors_on = pin->GetOrAddBoolean("floors", "on", floors_on_default);
    if (!floors_on) {
        pin->SetBoolean("floors", "disable_call", true);
    }

[c-prather]

This isn't really a supported configuration, since FOFC uses the floors to determine what zones to knock back to first order, as you've noticed. We should just error if these are specified together...

Disabling the call is an option, but if you're injecting floors in the normal frame then they will still be applied as a part of the primitive variable inversion, inverter/inverter.cpp.

Generally, any simulation which uses FOFC should apply the floors as a fallback, so I'm not inclined ever to support this. Evolving the magnetic field and total energy separately will always require applying a floor to fix negative-temperature zones.

[phyzan]

Yes indeed, that combination of parameters did not make sense. A simple runtime error sounds right, just to avoid segfaults (or worse, silent UB).