Telcon: 2023 06 21

Wednesday June 21st, 9am PT (UTC -7:00)

Attendees

• Peter Scheibel (host)
• Tammy Dahlgren
• Mark Krentel
• Massimiliano Culpo
• Chris Green
• Richarda Butler
• Vivek Sharma
• Wyatt Spear
• Jakov Petrina

Agenda

(this meeting is just for Q&A, there are no pre-planned topics)

• Chris: on unwanted rebuilds
  • Referring to prior meeting where building successive environments appears to rebuild (or build unexpectedly-distinct versions of) packages/dependencies
  • New info: the concretizer is deciding it can do something different for build-only dependencies
  • Builds a compiler which depends on zstd
  • Then attempt to build another env with a matrix
    • unify_when_possible and reuse on in this env
    • expecting higher-level deps to build distinct instances
    • There are lower-level deps also having alternative versions built though
    • In particular depending on a distinct version of zstd (it does not reuse the older zstd built for the compiler in the first env)
  • spack solve --show can include more info on generated facts
  • Latest information (including build with zstd version pegged to match pre-installed package) added to https://github.com/spack/spack/issues/37781.
• Vivek: ping on https://github.com/spack/spack/pull/36800
• Wyatt: sanity check for compilers/externals
  • spack compiler find is generating compiler entries that don't include modules but need them to run
    • this is on a redhat system
  • spack audit could be modified to verify that compilers are still valid (e.g. that their binaries still exist)
• Wyatt: loading any PrgEnv reloads darshan
  • Some packages won't build with darshan loaded (e.g. tau)
  • Can tau's package install method unload darshan?
    • e.g. taus install method could call unload(darshan)
  • Mark: setup_build_environment could unset the variable that controls darshan activity
    • however, tau doesn't necessarily check this variable and in fact checks other env variables like the set of loaded modules
  • Side note: darshan imposes compiler wrappers, and it is anticipated that the spack compiler wrappers in turn override those