add specs for conduit tests

[cyrush]

No description provided.

[adrienbernede]

@cyrush For the duration of your PR, you should change ENV_NAME to be conduit in .gitlab-ci.yml.
I am actively trying to find a better way.

[adrienbernede]

Right now, it triggered a pipeline for the radiuss environment, because that’s the default.

[cyrush]

A few notes:

The system configs (packages.yaml) are based off of @davidbeckingsale's - however they are not exactly the same b/c:

I need to test building conduit's python support -- and since it extends python, spack has to build python as part of the process.

[adrienbernede]

One way we could "uniformize" this would be to agree on a limit for things to share, and things that should be in the project spack.yaml:

• We share CMake -> packages.yaml (shared)

• We share compilers -> compilers.yaml (shared)

• We share Cuda -> packages.yaml (shared)

• We share MPI -> packags.yaml (shared)

• We don’t share python -> spack.yaml (project specific)

Then the questions is: what about small utilities (m4, pkg-config, and such)?

[adrienbernede]

That’s because we can have two source for the packages: section.

[adrienbernede]

Note: clingo compiler takes ages, that’s what we are looking for at the moment.

[cyrush]

this is where my head starts to spin - I feel like we get into a scenario where we need scripts to combine (override, union) shared packages. Solving this in a nice way would make sharing all of these things much much easier.

[adrienbernede]

If only we knew the nice way :)
The only way I see is to define rules. But maybe I just have blinders!

[davidbeckingsale]

Does spack configs allow this kind of "overriding" with multiple environments? A bit like how you can have repos that override packages...

[adrienbernede]

@davidbeckingsale my impression is that we can override part of the configs directly in the spack.yaml:

• We include the shared config as a starting point.
• We define locally to the spack.yaml what we want that is specific.

I don't know if there is such a thing as stackable environments. All I know is that environment modularity is currently being revisited by @becker33

[adrienbernede]

@cyrush I hope the lassen jobs are producing useful input (there was a failure of one conduit spec).

In the meantime, I am trying to decrypt the "KeyError" on quartz.

[adrienbernede]

Believe it on not but it looks like just today there were 2 changes in spack@develop that "alter" things in our pipelines.
It is creating a combo error that is tough to debug, but I was in touch with @scottwittenburg and we identified a path forward to find where the issue comes from exactly.