Questions for Spack meeting

Can we get some more info about this issue?

Seems like it has been acknowledged as something to be fixed?

As a work-around to above I’ve tried to use the spack hash to force package reuse as recommended in a previous telecom, but this does not seem to work

See

Can a variant support an arbitrary string as user input? e.g. variant(var, default="none", description=var, values="*", multi=False)

Answer: Dynamic variants: we don’t seem to document this anywhere I can find, and I don’t see any examples in our builtin repo, but we do have a test which basically does what you are doing in spack-packages/packages/um/package.py at 9e984d69c91580748fc05b5af46dd5a9566f565a Ā· ACCESS-NRI/spack-packages Ā· GitHub, so we are enforcing it at the test level (FWIW that test is called test_wild_card_valued_variants_equivalent_to_str)

Can the compiler wrapper output be put in the .spack directory when a standard spack installis executed? We are using Spack v0.22 and we need to run spack -d install and the output (spack-cc*.out) files are in the working directory.

Can Spack provide a command to cd into a package’s build log directory? e.g. spack -e access-esm1p6-pr94-14 cd --logs um7

Are there any ways to enforce a single compiler down a dependency chain (for example, with spack.packages.all.require: [%gcc@13.2.0]), but also allow good reuse of packages?

Some attempts:

spack:
  specs:
  # In which the below first-level deps use the gcc@13.2.0 compiler
  - cable ^cmake/bbw5qer ^gmake/nql6rht ^netcdf-c/o343qzy ^openmpi/mglfhd

This led to most dependencies down the chain using %gcc (or a version of it), but not all.

Should an SPR define url or url_for_version? Would Spack accept an SPR with neither into the upstream repository?

Can the behaviour of vcheck in the compiler wrapper be changed? (Do not use -V in fflags: it fools the Spack compiler wrapper by penguian Ā· Pull Request #315 Ā· ACCESS-NRI/spack-packages Ā· GitHub)

Can Spack be configured to use an external module file instead of generating its own?

1 Like

In terms of provenance/build recreation/disaster recovery for builds in spack 1.0, is the following all that is required to rebuild exactly?

  • The spack.lock file
  • The git hash of spack used to build the above lockfile
  • The git hash of spack-config used to modify spacks configuration for ACCESS-NRI’s instances
  • The git hash of spacks spack-packages repository
  • The git hash of ACCESS-NRI’s spack-packages repository