-
Notifications
You must be signed in to change notification settings - Fork 67
CCPP Framework Meeting Minutes 2026 06 04
Courtney Peverley edited this page Jun 4, 2026
·
5 revisions
Attendees: Michael Kavulich, Jesse Nusbaumer, Dustin Swales, RhaeSung Kim, Courtney Peverley, Dom Heinzeller, Cheryl Craig
GitHub issues, PRs, discussions
-
New test to demonstrate GNU failure #689 - worth adding as additional test in capgen-ng?
- Associated issue: Unallocated fields passed into Group Caps #690 will be closed by #759
-
DRAFT FOR DOM: Feature/capgen_for_fuji #718now CLOSED - DRAFT Update to capgen-ng "sneak preview"
capgen-ng progress
- Dom updated issue #759
- New test for auto-clone constituents
- Flag to enable auto-clone constituents (in metadata) for use by CAM-SIMA
- NEPTUNE complete
- CAM-SIMA python unit tests pass; kessler test passes with GNU
- Had some tricky constituent mods
- Added capgen compatibility layer to handle things CAM-SIMA does that framework tests didn’t do
- CCPP-SCM all suites compile and all manual tests runs
- Dustin made update to get single precision builds to work
- CAM-SIMA compatibility layer
- Goal is to slowly phase out the compatibility layer
- CAM-SIMA talks to top-level python interfaces in capgen, also gets back internal objects from capgen and uses parse tools, etc
- Problematic to use non-”public” interfaces
- SIMA would ultimately only like to interact with public interfaces
- Can pull the fortran writer over into SIMA
- Fortran writer lives in CAM-SIMA code base (but CCPP caps are still in fortran)
- Changes to CAM-SIMA physics: renaming “finalize” to “final”; constituents can’t be intent(out) -> change to intent(inout)
- Longer term differences: horizontal_loop_extent => horizontal_dimension; no auto-clone constituents (defined only in metadata)
- Long-term language interoperability?
- We’d need a new interface to a new language; Dom recommends crossing the bridge when we get there
- Command-line interface vs python interface
- capgen-ng has command-line, would we rather have python interface long-term?
- Command line lends itself to cmake build system (used by non-SIMA models)
- After Dom looked at CIME stuff, would like to build python interface as well (with feature parity - in/out should be equivalent)
- Dom thinks it’d be pretty quick to create the python interfaces
- ccpp-physics meeting recap
- Discussion was about physics cleanup; using suite / interstitial variables
- In original capgen, if variable is not provided by the host model, it is handled by the caps (possibly promoted); new capgen has simplified version - everything lives in ccpp suite data module, defined at suite level
- Everything is now allocated once
- Potential problem with physics scheme itself wanting to allocate something? TBD.
- Dom open to new name suggestions for this new version of caggen