Norma does not work correctly with Julia-1.12

[ikalash]

Last week, @alejandro-n-diaz found that Norma does not work correctly with Julia-1.12 on his Mac, and I have just confirmed this in RHEL. The attached .txt file shows the error that are obtained when running tests. There seem to be a variety of errors, including those related to bounds checking and threading.

tests_out.txt

[lxmota]

As a temporary workaround run like this on Julia 1.12:

julia -t 1

There is an issue with allocating threads at run time vs at compile time that needs to be fixed.

[lxmota]

@ikalash @eparish1 I have a fix for this but all the opinf tests fail with it. I don't know if that is because of the fix or because of Julia 1.12. Check the branch julia-1-12-threads and run with Julia 1.12 to verify. In addition, the test nnopinf-schwarz-overlap-cuboid-hex8.jl requires pytorch. We should avoid any Python dependencies.

[ikalash]

That's interesting. What are the errors for the opinf tests?

[lxmota]

Here is the log file for the --quick option. The comparisons to the baseline quantities fail.

test.log

I am going to merge this branch to main and disable the failing tests for the time being so I can move forward.

[ikalash]

That's interesting. Some of these differences are nontrivial. This is even if you run with 1 thread?

[lxmota]

Yes, these were run on one single thread as before.

[ikalash]

I saw you disabled the OpInf tests. I would prefer not to merge the 12.1 fix until we get the OpInf stuff fixed so that the tests can run. I worry that if the tests are not run, they will get broken.

[lxmota]

I understand but since I'm leaving soon and bound to forget, I'll go ahead as planned. The tests are easy to enable again and they remain active in the julia-1-12-threads branch.