Skip to content
Peter Scheibel edited this page Feb 7, 2024 · 24 revisions

Wednesday February 7th, 9am PT (UTC -8:00)

Attendees

  • Peter Scheibel (host)
  • Mark Krentel
  • Wileam Phan
  • Massimiliano Culpo
  • Davide DelVento
  • Jakov Petrina
  • Nick Romero

Agenda

There are no pre-planned agenda items: this meeting is just for Q&A

  • Mark: ping on new hpctoolkit PR: #42513
  • Davide: does anyone install jupyter-lab with Spack?
    • Install succeeds, but interactive plots don't work
      • Complaints about not finding javascript extensions
    • Note: jupyter-lab is installed w/ Spack, kernels are installed w/ pip (by end users)
    • Spack version is 0.18
    • Wileam: suggest installing w/ backend=agg,nbagg (default is agg)
    • Wileam: points to py-matplotlib-inline adding new dependencies in the last few months
    • Davide: will try with newer Spack
  • Davide: jupyter-lab startup prints warnings for all py-* dependencies (possibly related):
    ==> Warning: couldn't get environment settings for py-mako@1.1.6 /jdfvhms
    Invalid version specifier
    
  • Nick - example scenario: I do spack foo%gcc and then need to do spack foo%oneapi is there anyway to tell Spack not to rebuild the build dependencies with OneAPI and just used the ones build dependencies with GCC (i.e. only rebuild link dependencies with OneAPI).
    • Currently externals is the best way to do this: set buildable: false
      • This is commonly used for build dependencies
      • You can also build some/all of these with Spack and refer to them as externals with Spack
        • And you can use the same Spack instance (use one env to make CMake etc. and then add those as externals in a new env)
    • Massimiliano: there is an experimental concretizer option (see spack config blame concretizer, duplicates: strategy: full)
    • See also (more as an item of interest than something necessary to achieve this particular goal): PR to bootstrap glibc which is uses a similar multi-environment strategy
Clone this wiki locally