fix(deps): specsolve installs a polars older than 2.0, on which a row's dual can come back empty and a model can take nine times the memory

[FBumann]

Prompt: Do a pr that fixes the polars 2 issue. Cap it for now!

Note

The following content was generated by AI.

specsolve now requires polars>=1.30,<2. On polars 2.0.0, released on 2026-10-06, two tests get an empty dual where a row has one, and storage/s takes about nine times the memory. Making specsolve work on polars 2 is not part of this PR.

What fails on polars 2.0.0

Both results come from the same tree, with only polars swapped (1.44.2 against 2.0.0):

• Suite. Two tests fail: dual('budget_row') returns 0 rows where 1 is expected.

  • tests/test_resolution_parity.py::test_the_empty_coordinate_builds_on_both_lanes
  • tests/test_resolution_parity.py::test_a_masked_scalar_variable_takes_its_row_with_it[masked-in]

  This is how CI went red on docs: the published benchmarks are re-measured on specsolve 0.6.0 #1850, and it fails the same way on every PR and on main.

• Memory. test_emit[storage-s-specsolve-highs] with --benchmark-memory, three rounds of one local run:

polars	peak_bytes	RSS
1.44.2	119–123 MB	234 MB
2.0.0	1091–1146 MB	1072 MB

CodSpeed reported the same direction on #1850: the storage/s specsolve cells went from about 17 MB to about 776 MB.

What changed, and gates

• pyproject.toml. polars>=1.30 becomes polars>=1.30,<2. tests/test_tooling_pins.py already accepts a ,< ceiling. The floor of 1.30, and the floors environment that pins it, do not change.
• uv.lock. Relocked with uv lock. Besides the polars specifier, the relock brings in three pins that pyproject.toml on main already declares and the lock had not caught up with: ruff 0.16.9, zensical 0.0.66 and pymdown-extensions 12.1. I did not edit the lock by hand to keep them out.
• CHANGELOG.md. The line for this PR.
• The cap works. uv pip compile of polars>=1.30 resolves 2.0.0. With ,<2 it resolves 1.44.2. A fresh pixi environment in this worktree also resolves 1.44.2.
• Gates. pixi run check: 4996 passed, 564 skipped, 1 xfailed, the two tests above included. This was run in the fresh environment, which has polars 1.44.2.
• Not run. test-floors, because the floors pins do not change. docs-build, because no page changes.
• Not done. No test asserts the ceiling. With the ceiling removed, a fresh install resolves 2.0.0 and the suite fails on the two tests above, which is the guard. A test that pins <2 would have to be deleted with the cap.
• Type. fix(deps), as in fix(deps): an installed specsolve release keeps the mathspec minor version it was released against #1816: the ceiling changes what the package accepts.

🤖 Generated with Claude Code

https://claude.ai/code/session_018nKGXp8uKnMeJACjFYwbdY

---

Generated by Claude Code

[codspeed]

Merging this PR will not alter performance

✅ 67 untouched benchmarks
⏩ 307 skipped benchmarks¹

---

_{Comparing claude/polars-below-2 (8c7b794) with main (331a839)²}

Footnotes

1. 307 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. ↩

2. No successful run was found on main (aa90239) during the generation of this report, so 331a839 was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩