CCPP Framework Meeting Minutes 2025 03 13

Agenda

• GitHub issues/PRs
• New items
• Updates
• Other business
• notes

---

Attendees: Michael Kavulich, Dustin Swales, Dom Heinzeller, RhaeSung Kim, Jordan Powers, Cheryl Craig, Michael Waxmonsky, Jesse Nusbaumer, Courtney Peverley, Soren Rasmussen

GitHub issues/PRs

CCPP Framework (issues, PRs, discussions)

• Add support and testing for equivalent units #571

  • On the back-burner

• Allow for differing module/file names Issue#636

  • Capgen in SCM: Differing source and module name #637
    • Steve suggested a different approach: https://github.com/NCAR/ccpp-framework/pull/646

• Current intel compiler does not allow >255 subset arguments Issue#645

• Change default CLI behavior to newline separated output Discussion#647

• Host metadata file not flagging mis-named variables Issue#648

Drafts

• Testing refactor #612

  • Continued updates and discussion

• Capgen in SCM: DDT subfields into Group Caps #640

  • Ready for review?

Standard names (issues, PRs, discussions)

• Changes after rename of CCPPStandardNames --> ESMStandardNames #98

New items for discussion

Updates from last time

Previous notes

Other business

Meeting notes

CCPP Framework

• #571 - equivalent units
  • Dustin holding off hoping that he never has to implement the prebuild side (someday when prebuild is no longer used)
  • Dom - should be easy to fix prebuild side (just hardcode to skip)
    • Dustin to make it work eventually
• #636 - differing source and module name
  • Steve G suggested different approach - opened his own PR #646
  • Will use this approach instead & get rid of schemes that use old method in UFS (interstitials instead) such as https://github.com/NCAR/ccpp-physics/blob/main/physics/Interstitials/UFS_SCM_NEPTUNE/GFS_suite_interstitial_phys_reset.F90
  • Courtney and Dom to review; Dom will make necessary changes to UFS
  • Dom - had similar situation here - https://github.com/NOAA-EMC/fv3atm/blob/b604f35464e4de395072da115b137c67bf70ca03/ccpp/driver/CCPP_driver.F90#L143
    • Call reset routines (type-bound procedures) in host model instead
• #640 - DDT subfields in group caps will come out of draft after #646 is merged

• 255 subset arguments issue (#645)

  • Dustin found workaround by making multiple groups
    • Dom to take lead on splitting into groups
  • Courtney closed issue as “not going to fix” and we can reopen an issue if we run into this again
• Dustin running into bottlenecks on UFS physics/host/MPAS work
  • Grant to bring up in upcoming planning meeting
  • Dom to bring up in Friday meeting
• Change default CLI behavior to newline separated output (discussion #647)
  • General agreement, should make updating tests easier
• Host metadata file not flagging mis-named variables (#648)
  • Agreement that we’ll close it as a non-issue
  • Physics schemes only allow “scheme” or “ddt”, needed way to allow typedefs to be passed to schemes
    • Back to initial issue described in #636 discussion above (will call typebound procedure on host side)
• Dom - one of the problems w/ capgen is it’s so complicated
  • Maybe we try pointing a LLM at it to generate some documentation?
    • Michael is aware of a tool that will generate workflow documentation - will dig it up - https://gitdiagram.com/
  • Cheryl - would be nice to generate it and then pass it by Steve
  • Courtney to poke around for a tool
  • Dustin to help iterate on documentation while he’s “in it”
• Dom - UFS software architecture team meeting - presentation by Ed Hartnett who advocated for module-izing UFS and its components as much as possible He wants everything to be available as a module installed on its own and then made available to whatever you’re building; not possible in CCPP at the moment
  • Question: do we think it’s possible to come up with a design for CCPP where we can do that?
  • We at least want to be able to have justification for not doing it (or doing it?) to take back to Ed
  • Dustin - why? If it’s all about unit testing, it shouldn’t be necessary to do this library approach (just do more/better testing)
    • Unit testing is insufficient for testing our models
  • Jesse - even if we want to do it, it’d be so much work
    • Even when you CAN unit test, it can be hard to come up with non-trivial test cases
  • Dustin - physics scheme developers should be the ones doing this testing
  • Jesse - olive branch could be: trying to test discrete functions vs full schemes
    • We’ve been trying this in SIMA - https://github.com/ESCOMP/atmospheric_physics/pull/193