omega-h: Use seacas for exodus (instead of trilinos)

[Fuad-HH]

Trilinos was needed for exodus mesh files which only needs seacas. Now secas is directly used by Omega_h instead of through trilinos.

[cwsmith]

notes from offline discussion:

• add conflict for +trilinos +kokkos and +trilinos +seacas
• restrict +seacas dep for latest omegah tag scorec-v11.1.0

[bernhardkaindl]

Hi @cwsmith, @Fuad-HH and @Sichao25:

• I see that the git diff context of omega-h: add v11.2.0-scorec and a python variant #5792 has some lines this PR modifies. So if 5792 would be merged, this PR would need to be reworked.
• As @Sichao25's PR 5792 is new, when we merge this PR first, he can then easiliy rebase his PR on top of the merge in develop.

Is that fine with you?

If there are no objections, I'd approve and merge this PR tomorrow and add a note for @Sichao25 to rebase to #5792 on top of these changes (they just change context, so it should be relatively straight-forward form him to do it) after the merge.

The other option exists of course too, but someone, (@Sichao25) would have to do the rebase with conflict resoltion and add it to their PR or open a new PR.

This PR can't be easily rebased, at least GitHub says:

Rebase conflict between head and base.

When I try to use the Rebase option of the Update button of this PR. Update with merge was successful.

Later or tomorrow, we should see how this PR did in CI.

[cwsmith]

@bernhardkaindl Thanks for checking on this. Yes, lets merge this first then rebase #5792.

[Sichao25]

I'm fine with merging this first. Thanks for letting me know.

[bernhardkaindl]

@cwsmith, approving with your approval to make progress for the next PR!

@Fuad-HH:

When enabling +exodus, there where builds which failed with:

/usr/bin/ld: libomega_h.so: undefined reference to `nc_set_log_level'

However, other builds were successful.

I wasn't able to pinpoint the case yet, it must be a missing -lnetcdf-c or so when linking libomega_h.so.

Of course, it didn't show with ~shared, and the trigger seems to be:
spack install omega-h +exodus +mpi +shared

I didn't work this out completely, but with +mpi, I see parallel-netcdf-c is used and so far, I saw nc_set_log_level seems to be inly in the non-parallel netcdf:

grep -r nc_set_log_level ../spack/opt/spack/*/*netcdf*
../spack/opt/spack/linux-skylake/netcdf-c-4.10.0-oacmggfhgct2hgkvsf2el7os63aj7mok/include/netcdf.h:nc_set_log_level(int new_level);
grep: ../spack/opt/spack/linux-skylake/netcdf-c-4.10.0-oacmggfhgct2hgkvsf2el7os63aj7mok/bin/ncgen: binary file matches

@Fuad-HH, @cwsmith, as approved to base #5792 on top of this PR, so I'd say this the undefined undefined reference to `nc_set_log_level' can be solved that PR or another follow-up PR.

Cc: @Sichao25

[cwsmith]

Thank you @bernhardkaindl . We appreciate the help.

It seems this built against an older version of omegah as SCOREC/omega_h@4dd682e removed the call to nc_set_log_level.

[bernhardkaindl]

@cwsmith, this filter_file() removes (tested to build both affected versions) remaining debug call for the +exodus versions that still have the debug call:

Suggested change

	def patch(self):
	if "@:9.34.8" in self.spec:
	def patch(self):
	# https://github.com/SCOREC/omega_h/commit/4dd682ef16ebf2502239ad06883e9f10c611f1c4
	if self.spec.satisfies("@10.8.6-scorec:11.0.0-scorec"):
	filter_file(r"nc_set_log_level\(5\);", "", "src/Omega_h_exodus.cpp")
	if self.spec.satisfies("@:9.34.8"):

I guess a would be worth a suggestion to #5792 or alternatively a new PR.

Suggestion submitted: #5792 (comment)

BTW, off-topic: Is deprecating some of the older versions of omega-h a consideration for spack?