Telcon: 2025 07 01 - spack/spack GitHub Wiki

Tuesday July 1st, 5:00pm PT (UTC -7:00)

This is AEST 10:00am-11:00am (UTC +10:00) on Wednesday July 2nd, intended to be convenient for folks in that timezone

Attendees

• Peter Scheibel (host)
• Paul Leopardi
• Tommy Gatti
• Harshula Jayasuriya
• Aidan Heerdegen

Agenda

This meeting is for Q&A, there are no pre-planned topics (feel free to add one).

• Paul: anything deprecated for 1.0?
  • Peter: https://github.com/spack/spack/discussions/30634
    • I don't think anything at the command-line interfaces level
    • Spack library exposed to packages was constrained
• Tommy: for [email protected], is spack going to resolve that into a version
  • Peter: need to check on that
• Tommy: conflicting require sections:
  • If we have packages:require:all requiring compiler version
  • and a package wants a different compiler version
  • would there be a conflict?
  • Peter: no: packages:<specific-package>:require "throws out" packages:all:require
• Any issues with this approach?

class Roms(MakefilePackage):
     """ROMS is a free-surface, terrain-following,
     primitive equations ocean model widely used by
     the scientific community for a diverse range of applications"""
 
     homepage = "https://www.myroms.org/"
-    url = "https://github.com/myroms/roms/archive/refs/tags/roms-4.1.tar.gz"
+    git = "https://github.com/myroms/roms.git"
 
     version("4.1", sha256="cf25625066be3ea40fdd7bbe361f830d4415170636163b05bd338ac299809d4e")
     version("4.0", sha256="d14b4920e791ad24684f439c4751c2f1c38dbf9b82aa0d4d57def93e50a5a747")
@@ -51,6 +47,9 @@ class Roms(MakefilePackage):
     # Note: you cannot set USE_OpenMP and USE_MPI at the same time
     conflicts("+mpi+openmp")
 
+    def url_for_version(self, version):
+        return "https://github.com/myroms/roms/archive/refs/tags/{0}.tar.gz".format(version)
+
     def _copy_arch_file(self, lib):
         """AOCC compiler takes gfortran's makefile as reference"""
         copy(

• Peter (regarding prior point): you shouldn't have to remove class level URL attr to refer to roms@git..., but you do need git attr
• Harshula: variants that take arbitrary strings
  • Does spack frown upon this?
  • Peter: no
  • Harshula: example at https://github.com/ACCESS-NRI/spack-packages/blob/9e984d69c91580748fc05b5af46dd5a9566f565a/packages/um/package.py#L108
  • Peter: will look up how this is "officially"
  • flags not getting inserted into spack compiler wrapper (or rather, spack compiler wrapper not being executed)
  • explicit step where mpif90 is called with just ".o" files
    • Peter: I thought we redirected ld
    • Harshula: makefile does fuse ld=lld, which sets an abspath for ld
  • See: https://github.com/ACCESS-NRI/spack-packages/issues/255 and more-specifically https://github.com/ACCESS-NRI/spack-packages/issues/255#issuecomment-3008936267
    • Not a behavior from the makefile, but rather the ifx compiler (-o with obj files triggers abspath call)
  • Peter: Sounds like something that Spack should be doing in the compiler wrapper
• Aiden: can you constrain a spec to be a specific hash with require:
  • They were rebuilding ensf frequently
  • (putting it in the speclist)
  • Peter: I don't think you can require: this now
  • Aidan: we were doing speclist and removing a require: version
    • Peter: you should be able to keep the version requirement
  • See: https://github.com/ACCESS-NRI/ACCESS-OM3/blob/80e44e7f140e26af739e5af9d181d8ed48a9af5f/spack.yaml#L15
• Tommy: you are splitting the packages into their own repository
  • So are their versions totally independent?
  • Peter: yes; you should be able to move the core/builtin to wherever
  • Peter: we have core/package-repo versioning, so can record incompatibilities and auto-detect those issues
  • Aidan: is there a notion of per-package versioning?
    • Peter: not right now: we can add that if needed
• Harshula: is there a stable link for the recurring meeting?