Open climbfuji opened 6 days ago
@DusanJovic-NOAA @SamuelTrahanNOAA I'd appreciate if you could take a look at this. If what I am saying above is correct, then I don't understand why you are not getting these double free or out of bounds errors. Thanks!
I've seen other problems in the past with uninitialized indices from not calling get_tracer_index.
The nto and nto2 should always be initialized to something. I'd say zero is a good value if no tracers of that type exist.
EDIT: This is incorrect. See comments that follow.
If you call get_tracer_index for them (as I think you should), then they get value NO_TRACERS = -99, which then, together with your label_dtend_tracer logic, means they are skipped (index + 100 --> if <2 ignore) ?
Yes, my apologies, you are right. The correct fix is to always call get_tracer_index for those variables.
See https://github.com/NOAA-EMC/fv3atm/blob/a9364591091c836984a40107729720705847c195/ccpp/data/GFS_typedefs.F90#L5181
Model%nto
andModel%nto2
are only initialized if the CPP directiveMULTI_GASES
is set. But these variables always exist (see lines 1490-1491) and therefore remain uninitialized.In lines 5398-5399, they are used to set up the
label_dtend_tracer
array ifldiag3d
is true. Sinceldiag3d
can be true withoutMULTI_GASES
being set, it looks like uninitialized values ofnto
andnto2
are passed to that routine. Unless I am not seeing it. Depending on the system you are on, this can have bad consequences. Example: on my laptop and on Nautilus (a DoD Penguin Linux system), they both always have value zero. That results in double free memory corruption errors later in theadd_dtend
calls. On Narwhal (a DoD Cray system), they have very large but different integer values that result in out of bounds errors. This is with NEPTUNE and not with the UFS, but if I read the code inGFS_typedefs
correctly, the bug seems to be there, too?