Add public Solver.from_file API for existing LP/MPS files

[danielelerede-oet]

Related to open-energy-transition/solver-benchmark#604 (cc @eantonini).

Note

The following technical summary and validation report were generated with AI assistance.

Changes proposed in this Pull Request

Adds Solver.from_file(name, problem_fn, options=...) as a public, non-deprecated entry point for solving existing LP/MPS files without constructing a linopy Model.

The constructor validates the solver's file-reading capability and input path, copies solver options, and prepares the existing file-based solve path. The backend reads the input during solve().

This PR also:

• Updates deprecation warnings to point file-based callers to the new API.
• Adds API documentation, an example and a release note.
• Tests LP/MPS inputs, string and Path arguments, input preservation, solver options and invalid inputs.
• Compares results with the deprecated API, including variable and constraint names not generated by linopy.

Existing backend restrictions and result semantics remain unchanged. Without a linopy Model, label-indexed primal and dual arrays may be empty; this limitation is documented.

Validation

• Automated tests with HiGHS: 24 passed.
• Gurobi 13.0.3 smoke tests through Solver.from_file: LP and MPS passed, including objective, native variable values and input preservation.
• Combined solver tests with --forked: 212 passed, 20 skipped, 4 failed. The remaining failures occur in existing tests that open a second Gurobi environment while the first remains active, which the single-use license rejects.
• mypy linopy/solvers.py: passed.

Checklist

• AI-generated content is marked (see AGENTS.md).
• Code changes are sufficiently documented; i.e. new functions contain docstrings and further explanations may be given in doc.
• Unit tests for new features were added (if applicable).
• A note for the release notes doc/release_notes.rst of the upcoming release is included.
• I consent to the release of this PR's code under the MIT license.

[codspeed]

Merging this PR will not alter performance

✅ 181 untouched benchmarks
⏩ 181 skipped benchmarks¹

---

_{Comparing open-energy-transition:feat/solver-from-file (df2e7f6) with master (1b2ea76)}

Footnotes

1. 181 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[MaykThewessen]

We tested head df2e7f6 using our existing pixi test environment, without installing dependencies: all 24 upstream tests passed in 0.25 s, covering 12 cases under both semantics modes.

The small HiGHS LP/MPS cases cover file-byte preservation, native solver access, options, status, comparison with the legacy path and external variable names. This API looks useful for replaying exported BirdFlow MPS files when diagnosing solver options or versions. We have not yet replayed a full BirdFlow MPS through it; other backends and repeated-run behavior remain untested.

For that workflow, the empty model-indexed primal/dual results without a linopy model are worth documenting prominently. Consumers need native values and labels before closing the solver, plus explicit mapping back to PyPSA components.

One possible follow-up: check the native readModel return status and test a malformed input file. The unchecked parser status is pre-existing behavior, not a regression introduced here.

Our assessment and reproduction details are published.

Note: AI-assisted (written with GPT-6).

168-hour BirdFlow MPS replay

Additional large-model replay check, using head df2e7f6f380ef0133d93adb6ff481c5dfc3f704a:

A sealed 168-hour BirdFlow production float64 LP contained approximately 2.58 million rows, 1.66 million columns and 4.99 million nonzeros. Production linearized unit commitment was applied explicitly; native variable types were preserved and verified continuous. HiGHS 1.15.1 used HiPO, 3 threads and crossover on.

Native direct and Solver.from_file solves both finished optimal with valid primal/dual results. Objective difference was approximately 6e-6. Independent row/bound/activity/stationarity residuals passed a 1e-6 absolute gate. MPS SHA-256 was unchanged. Small MPS coefficient/cost/RHS rounding perturbations passed explicit tolerances.

Native vectors were not identical: maximum generator dispatch difference was 3634 MW and storage state-of-charge difference 11.3 GWh. Raw nodal-balance dual differences stayed below 1.05e-7; other dual vectors differed. This does not establish identical operational outputs or exported prices. No API defect appeared; no speed or annual-validation claim is made.