[FabianHofmann]

Changes proposed in this Pull Request

Note

The following content was generated by AI.

Model.matrices used ravel()[active] on each constraint's sign, rhs and scaling in _row_block. When those fields are stored lazily as stride-0 np.broadcast_to views (a scalar sign or scaling), this builds full-size copies. It also copies when every row is active.

Changes in linopy/matrices.py:

• A broadcast scalar field (all strides zero, size > 0) stays a 0-d array through _RowBlock (sign, rhs, scaling).
• When all rows of a constraint are active, the field is taken as a raveled view (reshape(-1)) and the boolean-mask copy is skipped.
• _non_unit handles 0-d scaling. _stack multiplies by a 0-d row scaling directly instead of calling repeat_per_row, and assigns 0-d rhs/sign by broadcasting.
• The con_scaling concatenation broadcasts 0-d blocks to their row count.

The public Matrices fields and dtypes stay the same (sense is still a U1 array, and b and con_scaling are still full float arrays), so this is not a breaking change.

Tests in test/test_matrices.py:

• test_row_block_keeps_broadcast_fields_lazy: scalar sign and scaling stay 0-d, with all rows active and with some masked.
• test_matrices_scalar_and_array_fields (dense and sparse): covers scalar sign/scaling, masked rows, mixed signs and per-cell scaling. It compares A, b, sense, con_scaling and clabels against explicit expected arrays. The same expected arrays also pass against the unmodified master implementation.

Benchmark

SciGrid-DE tiled to 1.79M vars / 4.29M rows / 7.18M nnz, cloud VM, best of 5:

	before	after
matrix assembly (best)	1.01 s	0.63 s
matrix assembly (median)	1.08 s	0.81 s

In line profiles, con.sign.values.ravel()[active].astype("U1") alone took about 140 ms before this change.

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.

---

Generated by Claude Code

[codspeed]

Merging this PR will not alter performance

✅ 181 untouched benchmarks
⏩ 181 skipped benchmarks¹

---

_{Comparing perf/matrices-lazy-fields (b91d191) with master (6c832f2)}

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. ↩