Telcon: 2023 05 10

Held Wednesday May 10th, 9am PT (UTC -7:00)

Attendees

• Peter Scheibel (host)
• Jakov Petrina (Sartura)
• Jakov Smolic
• Mark Krentel
• Massimiliano Culpo
• Tammy Dahlgren
• Dom Heinzeller

Agenda

This meeting is just Q&A: there are no pre-planned general Spack topics

• Mark: PR merged for writing and printing versions foo@=1.0 (https://github.com/spack/spack/pull/36273)
  • This means "exactly 1.0", and not something like @1.0.0
  • What happens if I don't have an =? Like spack spec foo@1?
    • Massimiliano: foo@1 could be satisfied by any 1.x or 1.x.y
  • We should update https://github.com/spack/spack/discussions with a discussion of these changes
• Jakov: considering Spack as a component of a firmware build system
  • Current issue: packaging of libc
    • This should be handled with the compilers-as-dependencies work
    • Massimiliano: as part of this, dealing with "separate concretization of build dependencies"
      • i.e. right now any package can only appear at most once in our dependency dag: this can be overly constraining for some cases where a different build dependency instance is needed for different packages
  • Current issue: cross compilation
    • Also should be addressed by the compilers-as-dependencies work
  • Peter: Can you manually assemble diverse specs in the meantime?
    • i.e. with spack-python
    • One issue: our query model expects to find one instance if you query dependencies
      • Our current draft PR on separate concretization of build deps could be useful for this: https://github.com/spack/spack/pull/35739
• Jakov: can you build differently on different architectures?
  • Yes: you can condition actions inside of build methods (like install) on spec details like arch/os/target
• Brian: there is a significant duplication of CUDA/ROCm logic across several CachedCMakePackages, I am working on a PR to move this into the CachedCMakePackages