feat(language): a where may read a predicate through a relation with at()

[FBumann]

Prompt: Update lpspec to the latest mathspec release (119)

Note

The following content was generated by AI.

The pin moves from math-spec alpha.116 to alpha.119. Both lanes now build the new at(<predicate>, by=, over=, into=) in a where: (energy-models/mathspec#634), and they agree on the result. A coordinate with no relation row reads false. 70 insertions, 34 deletions.

What moved

• PulledBackPredicate: the relational lane walks the coordinates the operand admits through walk_join, as an expression's at does. The linopy lane reads the evaluated mask through operator_at. Before this change, both lanes hit assert_never.
• PiecewiseExpansionError is gone upstream (refactor(language): an expression resolves straight into the program's own nodes energy-models/mathspec#638). A piecewise block now raises DimensionError, so lps.PiecewiseExpansionError goes too, with no alias. test_api, test_architecture and test_piecewise follow, and so does docs/reference/api.md.
• ArithmeticComparison is gone upstream (refactor(program): a comparison of expressions is one node before and after lowering energy-models/mathspec#631). The two lanes' branches for it were never reached, so they go. NEVER_LOWERED in test_resolution_parity goes with them.
• Tests: test_a_where_reads_a_relation gains 2 cases and test_a_relation_where_agrees_with_the_oracle gains 3: a total relation, a partial one, a negation and a conjunction. COVERED_ELSEWHERE names the oracle test for PulledBackPredicate. Before the implementation, the coverage guard failed on it.
• Upstream now refuses sum() over a scalar, so the carried-parameter probe in test_strategy reads soc_initial bare. It still asserts the refusal names "carried".
• uv.lock is relocked. Only the math-spec entry changed.

Mutation table

Run by hand on the committed tree. Each file was restored through git checkout -- and __pycache__ was dropped. The runs cover test_label_coords.py and test_resolution_parity.py.

Mutation	Result
linopy: a missing relation row reads true (fillna(True))	caught, 1 failed
polars: a missing row reads true (fill_null(True))	caught, 5 failed
polars: the operand's mask is ignored (masked(..., None))	caught, 5 failed

Gate

ruff check .            clean
ruff format --check .   327 files already formatted
pyrefly check           0 errors (20 suppressed)
pytest -q -n auto       3796 passed, 403 skipped, 1 xfailed, 11 failed

The run used uv with the [linopy] extra, not pixi. The xfail is the known osemosys_utopia / #894 case. Of the 11 failures, one was the test_strategy probe, which is fixed in the second commit. The other 10 are the gurobi and xpress parametrizations of test_diagnostics and TestThePositionalHandoff. They raise ModuleNotFoundError because neither package is installed here, and they fail the same way on origin/main.

Not run: docs-build, docs-test, test-floors, test-bench, the gurobi and xpress sinks. I did not regenerate the gallery pages. The typesetter changed upstream (energy-models/mathspec#629), but the doc tests passed.

🤖 Generated with Claude Code

https://claude.ai/code/session_01LzN7uYEvHW9dDGCmtAzbkD

---

Generated by Claude Code

[read-the-docs-community]

Documentation build overview

📚 lpspec | 🛠️ Build #34716847 | 📁 Comparing ec3a5bb against latest (813aeb3)

  🔍 Preview build  

1 file changed

± reference/api/index.html

[codspeed]

Merging this PR will not alter performance

✅ 24 untouched benchmarks
⏩ 58 skipped benchmarks¹

---

_{Comparing claude/kind-allen-76xzst (ec3a5bb) with main (813aeb3)}

Footnotes

1. 58 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. ↩