Kernel crash with `mf6.finalize()`

[aufdenkampe]

I get a kernel crash when running mf6.finalize() in your example docs/examples/notebooks/Extensions.py or on my own.

The Jupyter log provides the following info:

[error] Disposing session as kernel process died ExitCode: 2, Reason: At line 84 of file ../src/Model/ModelUtilities/SpdisWorkArray.f90
Fortran runtime error: Attempt to DEALLOCATE unallocated 'this'

I'm running it all from custom conda environments, running Python 3.11 or 3.12, on MacOS (arm64), which installs:

modflow6                  6.6.1                h5773a5b_0    conda-forge
modflowapi                0.2.0              pyhecae5ae_1    conda-forge

Here's one of the conda environment files used: https://github.com/p-ortega/mf6rtm/blob/main/benchmark/environment.yml

[jlarsen-usgs]

@wpbonelli, @aufdenkampe

If I change block 39 from

mf6.finalize()

to

mf6.prepare_solve(model.solution_id)
mf6.solve(model.solution_id)
mf6.finalize_solve(model.solution_id)
mf6.finalize_time_step()
mf6.finalize()

then the kernel does not crash.

It looks like there is an allocation error on the MODFLOW (libmf6) side. My guess is that the X array is not allocated until we do a solve now, which is causing the crash. Not sure if we want to patch it on the notebook side, or if it should get fixed on the MODFLOW side. Any thoughts?

[aufdenkampe]

@jlarsen-usgs, thank you for providing a quick fix for the example and my current use cases!

That would be awesome if you could also fix it in the Fortran code (kernel crashes are never good), but I understand that might require a bit more time to tackle. Should I create a new issue for that, or should we keep this one open?

[wpbonelli]

Yeah, I think we want to patch MF6. I meant to take a look but haven't had a chance.

@jlarsen-usgs @aufdenkampe if you prefer to merge #80 in the meantime and close this, a new issue on MF6 would be good. Otherwise I guess we can just track it here?

[jlarsen-usgs]

@wpbonelli

I can merge these changes, so the example doesn't crash anymore. We can always revert them once the allocation issue has been taken care of on the MF6 side. Either way, we probably want to link this issue to the MF6 updates.

[wpbonelli]

MODFLOW-ORG/modflow6#2322 should fix this. I should have gotten this in for 6.6.2 yesterday, but didn't have the presence of mind. So for now the manual prepare/solve/finalize will still be necessary unless you're working with the mf6 nightly build.

[aufdenkampe]

Thank you for addressing this so quickly!