Telcon: 2023 03 22

To be held Wednesday March 22nd, 9am PT (UTC -7:00)

Attendees

• Massimiliano Culpo (host)
• Dom Heinzeller
• Mark Krentel
• Jack Morrison
• Wileam Phan
• Phil Reiger
• Jawad Ahmed

Agenda

This meeting will be Q&A: there are no pre-planned general Spack topics.

• Mark How can I make it easy for hpctoolkit users to build with configuration suggested by hpctoolkit maintainers?

  • Mark: Pointed to this PR https://github.com/spack/spack/issues/16909 where a feature is suggested to cover this use case. The core idea is to allow a package to override default values of variants etc. of its dependencies.
  • Massimiliano: Expressed concerns about the proposed approach (see https://github.com/spack/spack/issues/16909#issuecomment-1476969126 and following discussion).
  • Massimiliano: Proposed to investigate a solution where "suggested" configuration for a package can be stored alongside the package.py, and imported in an environment with e.g. a command. That would be an explicit action done by the user (which would thus be aware of changes in configuration), but considerably simpler than the current workflow.

• Ahmed Is it expected that concretizing an environment with unify:when_possible leads to duplicate packages?

  • Massimiliano: That situation might happen. The only option that would error if there are duplicate in the environment is unify:true. See https://spack.readthedocs.io/en/latest/environments.html#spec-concretization

• Ahmed What would be the best way to use python-extensions with an environment?

  • Massimiliano: If possible for your use case, the best way would be to use unify:true and setup a view. See https://spack.readthedocs.io/en/latest/environments.html#filesystem-views

• Wileam Pinged for a review of https://github.com/spack/spack/pull/35919 Asked what to do with the build of LLVM in that recipe.

  • Massimiliano: Possibly, the cleanest way to deal with that is to modify the llvm package too and have gcc depend on it when the amdgcn variant is activated.

• Massimiliano Advertized new requires directive coming when https://github.com/spack/spack/pull/36286 is merged

  • Mark: There might be cases in the future where one_of needs to be exposed to the requires directive (e.g. dyninst can compile with GCC or Clang)
  • Massimiliano: The mechanism is already there for packages.yaml. If we need one_of it would be easy to extend the directive API to expose the same functionality as in the configuration file.