access-om3-nuopc: depend on esmf with -fp-model precise

[harshula]

• Also, use when="%intel"

[harshula]

Hi @anton-seaice , You can put your notes from testing, here.

[harshula]

Fixes: ACCESS-NRI/ACCESS-OM3#9

[micaeljtoliveira]

LGTM

[anton-seaice]

I built access-om3 using this spack-packages and the ACCESS-OM3 environment @ 3-spack-yaml-versioning (ACCESS-NRI/ACCESS-OM3@e6ae976)

Using model-config-tests (@7670f572f1ec3cc3198ebd5b62e7905a2579c5aa) I tried the following

checkout MOM6-CICE6, 1deg_jra55do_ryf (@070493071c807e26071062ed14be795f150426de)
(COSIMA binary, "access-om3-69f719964a27e23deae3923d60af01a081cc3fc0_main-fkzzqnp")

• model-config-test -m checksum = passes (after setting initial checksum)
• model-config-test -m checksum_slow = passes

changed binary to

access-om3-git.2024.04.0_2024.04.0-ut3rlqfs7kapqgn6ksnf75ysbfgchi4k from access-nri build

• model-config-test -m checksum = fail (checksums have changed - no surprise)

copy new checksum to testing/checksum

• model-config-tests -m "checksum or checksum_slow" = passes

git switch back to spack-packages@main

spack uninstall --dependents esmf
spack concretise -f 
spack install

changed binary to new build (access-om3-git.2024.04.0_2024.04.0-wiv6b2ibiwurjkghhactrpauzcmjmyx5)

• model-config-test -m checksum = passes
• model-config-test -m checksum_slow = passes

Based on COSIMA/access-om3#40, I expected the runs without this PR (i.e. spack-packages@main) to fail the checksum test. Any thoughts @micaeljtoliveira @dougiesquire ?

Maybe I didn't build properly? Maybe the test doesn't cover how the bfb repro issue was identified initially?

[anton-seaice]

I ran the 1deg_jra55do_ryf twice for one day runs, using the build without this change. The md5sum of the cice & mom6 restart files and the diff the restart files from the two runs show they are identical. So basically I can't reproduce the original problem. I wonder if it got fixed somewhere else unrelated to setting this flag?

[anton-seaice]

The restarts are reproducible, so lets merge this. Its possible we don't need to set the flags, but also I doubt there is a downside? So lets go with it