Telcon: 2021 08 18

Meeting to be held at 9am PST August 18, 2021

Attendees:

• Todd Gamblin
• Peter Scheibel
• Greg Becker
• Tamara Dahlgren
• Howard Pritchard

Agenda:

• Spack's compiler wrappers require bash, to support them on systems without bash we are considering either:
  • Making the wrappers posix-sh compliant*
  • Replacing the script-based wrappers with executables (compiled from C)

Possible agenda items for next week

• Spack compiler flag specification

  • Specifying compiler flags is hard now
  • We have flag_handler methods
  • Should we model specific flags (i.e. those which should be propagated vs. not)
  • We should also designate between cases where we want to add flags vs. set them
  • Currently flag_handler returns 3 values (which flags are added in the compiler wrapper, which are exported as shell vars, and which are fed to the build system)

• Packages which require multiple build systems: how to handle them

  • E.g. a package where earlier versions use make and later versions use cmake
  • Some packages will want to use a different build system on Windows
  • Side note: packages are only used at build time
  • ? Do we use when decorators on methods other than to constrain a method to only run for particular versions

• Defining a standard for shared and static libs

  • For some packages +shared might build only shared libs, or both shared and static libs
  • ~shared definitely means don't build shared

• Use variants

* note from @haampie: we rely on arrays & heredocs, that's tough in sh. @alalazo pointed out we can generate the sh script from a template, and hard-code the case statements instead of using loops and arrays. Might as well generate a C-script then, compile it, and have the lowest overhead compiler wrapper ;) https://github.com/spack/spack/pull/25380#issuecomment-901049740