Raise the SOCRATES and AGNI pins for GCC 16 symbol export

[egpbos]

Description

Raises two pins that must move together:

• SOCRATES 43b86d4d (2603.8) to c3296586 (2603.10), the current SOCRATES main.
• AGNI c7ffdf00 (1.12.3) to f94940b6 (1.12.4), the current AGNI main.

GCC 16.2 gives a PRIVATE Fortran module procedure hidden ELF visibility even where its bind(C) label grants external linkage, so dlsym cannot resolve it. The generated Julia wrapper modules and julia/src/SOCRATES_C.f90 each declared their contents private, which left 125 of the 128 PS_ entry points local in libSOCRATES_C.so and stopped AGNI at PS_create_StrDim. SOCRATES 2603.10 declares them public (FormingWorlds/SOCRATES#30). The previous pin carried only the earlier fix for real_kind_bytes (FormingWorlds/SOCRATES#29), which cleared the precision probe but none of the wrapper or radiative transfer entry points.

The SOCRATES pin cannot move on its own. AGNI compared SOCRATES versions by parsing the whole version string as a float, and 2603.10 is the first SOCRATES release whose minor number reaches two digits, so "2603.10" read as 2603.1: below the declared floor of 2603.6, and numerically identical to the real earlier release 2603.1. AGNI reported the library as out of date and setup failed. AGNI 1.12.4 splits the version string and compares major and minor as integers, and raises the floor to 2603.9.

The module version table in docs/Reference/module_versions.md is regenerated for both pins, which the generate_version_badges.py --check job requires.

No linked issue exists for this.

Bumping the SOCRATES pin invalidates the cached CI SOCRATES build, whose key derives from [tool.proteus.modules].

Validation of changes

Test configuration: macOS 15 arm64 (Darwin 25.6.0), GCC 16.2.0 (Homebrew), netCDF-Fortran 4.6.4, Julia 1.12.7, Python 3.12.14, both modules checked out at the refs pinned by this branch, SOCRATES cloned fresh and compiled with ./configure && ./build_code followed by the Julia wrapper build.

Symbol export, measured with nm on julia/lib/libSOCRATES_C.so:

Build	Exported (T)	Local (t)
New pin c3296586	128	0
Previous pin 43b86d4d, same tree and compiler	3	125

The second row is a control: only the two wrapper source files were reverted, everything else held fixed. At the previous pin PS_create_StrDim, PS_radiance_calc and PS_read_spectrum are all local, and PS_real_kind_bytes is the single exported entry point. Restoring the new pin and rebuilding returns the count to 128 exported and 0 local. The compiler is therefore ruled out as the variable.

Coupled behaviour, pytest tests/integration/test_integration_agni_transparent_limit.py tests/integration/test_integration_agni_greygas_interior.py with PROTEUS_CI_NIGHTLY=1:

Test	Result
test_transparent_greygas_olr_equals_blackbody_emission	PASSED
test_transparent_greygas_olr_scales_with_surface_emissivity	PASSED
test_agni_greygas_dummy_interior_coupling	PASSED
test_transparent_banded_olr_recovers_blackbody_emission	FAILED, pre-existing, see below

All four tests reach AGNI.atmosphere.setup!, which allocates SOCRATES.StrDim() and so calls PS_create_StrDim, the exact symbol that could not be resolved before the fix. Against the previous SOCRATES pin these tests fail at that call. The banded test additionally exercises the full spectral-file path through the SOCRATES two-stream solver, which the grey-gas tests do not touch.

Unit tier, pytest -m "unit and not skip and not slow and not integration" --ignore=tests/examples: 2987 passed, 41 skipped (optional dependencies absent), 0 failed.

Lint, structure and docs: ruff check src/ tests/ clean, ruff format --check src/ tests/ reports 343 files already formatted, bash tools/validate_test_structure.sh passes, python tools/generate_version_badges.py --check reports the table up to date.

Pre-existing failure in the banded test

test_transparent_banded_olr_recovers_blackbody_emission fails, and it fails identically without this change. Running it against the pins currently on main (AGNI c7ffdf00 with a SOCRATES build of the same vintage) produces the same value to the last digit, 287271.5671661585, so neither pin in this pull request moves the result.

The observed outgoing longwave flux sits slightly above the Stefan-Boltzmann limit rather than below it:

T_surf	Obtained F_olr	sigma T^4	Excess	Relative
1000 K	56719.7053	56703.7442	+15.9611	+2.82e-4
1500 K	287271.5672	287062.7050	+208.8622	+7.28e-4

The test tolerance is 5e-4, so 1000 K passes and 1500 K does not. Worth noting for whoever picks this up: the test also asserts F_olr < sigma T^4 on the grounds that band truncation and midpoint evaluation can only lose flux against the exact integral, and the measured values violate that in the same direction at both temperatures. That points at something other than tolerance calibration.

This is out of scope here. The SOCRATES change cannot be responsible: the diff from the previously pinned SOCRATES to c3296586 touches only julia/src wrapper files, docs and the version file, leaving the radiative transfer core untouched. Integration tests run nightly rather than on pull requests, so this does not gate the checks on this branch.

Checklist

• I have followed the contributing guidelines
• My code follows the style guidelines of this project
• I have performed a self-review of my code
• My changes generate no new warnings or errors
• I have checked that the tests still pass on my computer, with the one pre-existing failure described above
• I have updated the docs, as appropriate: docs/Reference/module_versions.md regenerated
• I have added tests for these changes, as appropriate: the existing AGNI integration tests already cover the affected path, and the symbol export itself is a property of the SOCRATES build rather than of PROTEUS
• I have checked that all dependencies have been updated, as required

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 93.46%. Comparing base (e2fd794) to head (4dc8266).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #863   +/-   ##
=======================================
  Coverage   93.45%   93.46%           
=======================================
  Files         113      113           
  Lines       16797    16797           
  Branches     2992     2992           
=======================================
+ Hits        15698    15699    +1     
+ Misses       1093     1092    -1     
  Partials        6        6           

Flag	Coverage Δ
unit-tests	87.25% <ø> (+0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[egpbos]

@nichollsh Tests on this fail because AGNI declares SOCVER_minimum as a float and so it sees 2603.10 as smaller than the declared minimum of 2603.6. How shall we fix this?

[nichollsh]

I noticed this last night, actually, and implemented a fix in the most recent AGNI commit: nichollsh/AGNI@f94940b

It should work fine if you pull changes to your local AGNI repo and update the hash pinned by PROTEUS here.

[egpbos]

@nichollsh Thanks, that works. Pinned AGNI at f94940b6 here alongside the SOCRATES bump to 2603.10, and the version check now accepts it.

Three of the four AGNI integration tests pass. The fourth, test_transparent_banded_olr_recovers_blackbody_emission, fails, but it fails identically against the pins currently on main, giving the same F_olr to the last digit (287271.5671661585), so neither bump causes it. Is this something you were aware of?

[nichollsh]

Interesting. One of my recent PRs did introduce some changes to ensure that the interior radius/gravity are robustly passed to AGNI from the helpfile. So maybe the blackbody calculation uses a different radius (in the test) compared to what AGNI is being provided with? This would lead to different fluxes.

[timlichtenberg]

Sorry for the churn here, Patrick. The conflict is my doing: the banded-OLR nightly fix (#866) bumped AGNI on main to 8a494d7c while this PR was open.

I merged current main in and reconciled the two pins:

• SOCRATES moves to 2603.10 (c3296586), as you had it. This is the change that matters for the GCC 16 symbol export.
• AGNI stays at main's 8a494d7c. That commit is newer than the 1.12.4 (f94940b6) this PR targeted and already contains it, so the two pins still move together (AGNI is well past 1.12.4). Taking the older pin back would have reverted the banded-OLR fix and turned the nightly red again.
• Regenerated the version badges to match.

Linux CI is green: SOCRATES 2603.10 builds and links against AGNI 8a494d7c and the full suite passes. The macOS tier failed earlier on a Zenodo outage and is green again on the re-run. Ready to merge from your side whenever suits.

[timlichtenberg]

Reconciled onto current main. SOCRATES 2603.10 is the GCC 16 symbol-export fix; I kept main's AGNI pin (8a494d7c) since it is already past 1.12.4 and includes the banded-OLR fix. CI is green. Approving to unblock the merge.