Skip to content

Migrate to ANYfeSolver 0.5.0 and ANYfeSolver-results 2.2.0 - #70

Draft
audunarn wants to merge 17 commits into
mainfrom
release/anyfesolver-0.5.0
Draft

audunarn wants to merge 17 commits into
mainfrom
release/anyfesolver-0.5.0

Conversation

@audunarn

@audunarn audunarn commented Nov 11, 2026 •

Copy link
Copy Markdown
Owner

Rename the solver to ANYfeSolver 0.5.0 and the independently installable NumPy results package to ANYfeSolver-results 2.2.0. The candidate starts at e0c5f32ab on main. It preserves historical archive identifiers, maps supported restart class names and retains exact-build and tamper checks. The solver uses the published SESTRA descriptor fix in ANYfileio 0.3.3.

The stable release is on hold. NLG-007 requires an imperfect plastic panel to complete arc-length analysis. Its local dent warps a plastic Q4 element that the current qualified formulation must refuse. The existing benchmark and refusal criteria both apply. The previously published migration previews are ANYfeSolver 0.5.0a1 and ANYfeSolver-results 2.2.0a1.

At f81375f, the checkpoint repair reproduces the archived stiffness of the two-patch fibre and curved multifacet S3 models. Both complete records replay exactly on Windows with OpenBLAS configured for Haswell, which reproduced the hosted mismatch, and with the original matching SkylakeX arithmetic. A wrong-record replay restores the prior owner state. The full current matrix is running on the candidate. The matched238 structural equilibrium gate remains open.

At e23aa8c4b, both wheels and source distributions passed the release artifact verifier, including separate wheel rebuilds from the source distributions. The Python 3.13 upgrade rehearsal upgraded the three previous public applications to the four local candidate wheels. The consumer entry points, pip check, solver CLI and V1/V2 replay passed after removal of the old package names. The earlier installed Windows native check passed 29 tests, including tamper refusal and fallback. The distributions compared at that revision contained identical runtime sources. The Windows and Ubuntu installed contact workflow passed at f54e3cde5. Cross-host contact checks · Current portable test matrix

The release remains held. On the historical 238-coordinate model, the current solver failed the first equilibrium predicate at the original tolerance of 1e-6. The relative force and moment errors were 0.006604931713040884 and 0.0000655082900502955. The six dependent checks were unrun. The earlier passing assessment is preserved. Warped Q4 verification and exact checkpoint replay also remain open. Both renamed distributions remain unpublished.

Final hosted results at e23aa8c4b were 40 passed and 9 failed in the 49-job Tests matrix. Two jobs reached the 2,400-second limit. The others cover beam/shell verification, protected S3 displacement hashes, two saved-checkpoint bindings, a warped Q4 gap, the D1 slide oracle and the machine-bound matched238 input. Both hosts passed their independent installed contact checks. Their comparison failed one of 29 primary quantities, the normalized equilibrium error for the six-triangle S3 V2D orientation case. Both current ANYstructure and ANYfem development migrations passed their applicable hosted checks. Solver matrix · Installed contact comparison · ANYfem · ANYstructure

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant