fix(data): a relation whose two columns are over one dimension builds

[FBumann]

Prompt: Then let's fix the bug first

Note

The following content was generated by AI.

Closes #1652. A self-map — rep_of: {columns: {snapshot: snapshot, rep: snapshot}, key: snapshot}, the language's own example of the form — passed check and then raised: polars' DuplicateError relationally, linopy's coordinate mismatch eagerly, and on sum(by=) the two lanes disagreed, one raising where the other built.

The walk named its value column after the dimension it lands on, which for a self-map is the name the key column already holds. The mapping table names its columns by side now — __walk key__, __walk value {relation}__ — and the join names its sides (left_on / right_on), so the landing is aliased to its dimension at the end rather than colliding at the start. On the eager lane the pullback puts the fine labels back as the coordinate: a row lands at the snapshot that read a value, not at the one it read.

What the probe says, before and after

	lps.check	relational	eager
at(price, by=rep_of, over=rep) before	passes	DuplicateError	ValueError: Coordinate mismatch
sum(p, by=rep_of, over=snapshot, into=rep) before	passes	DuplicateError	builds, unchecked
both, after	passes	builds	builds, and the two agree

Mutation table

Baseline for this environment: 10 failures, all gurobi/xpress not installed.

mutation	result
the one-value-column-per-relation guard (compiler.py:703-706)	caught
the walk selects its landing by the dimension's own name (compiler.py:845)	caught — 218 failed
the eager pullback keeps the labels it read (operators.py:120)	caught — 11 failed, 1 beyond the baseline
the join renames the mapping onto the fragment's dims instead of naming both sides (compiler.py:846)	not caught — recorded rather than hidden: it is an equivalent implementation, since what the bug needed was the landing to keep a name of its own, which the row above is the guard for

The first through python -m tools.mutate, which reported the tree clean afterwards; the rest by hand (changed lines, which the tool does not express), each with the same three precautions.

Verified

ruff check · ruff format --check · pyrefly check · pytest -q -n 4 — the suite reaches the same 10 failures as main in this environment, all a solver that is not installed. Run in a uv venv rather than pixi (no pixi here).

Not run: pixi run test-floors and test-bench — no pixi environment here.

The bug was reproduced first: tests/test_self_map.py landed as two strict xfails naming #1652 (commit 0324845), which XPASS on the fix and lose their markers in the commit that makes them pass. What was wrong is in their docstrings. The where: case and the duplicate-key refusal pass beside them, which is what said the data path was never the problem.

Coverage that moved: test_sum_by_relations.test_a_hand_built_walk_onto_two_columns_is_refused met a zip crash inside the mapping builder; the guard it probes is now a named assertion that fires before any relation is read, and the test matches that.

What I did not do

• The other three shapes stay refused — a relation whose key determines several value columns is refused, and the language walks it to both at once #1654, a bare relation is refused, and the language sums through one #1655, a shift, sum_back or position grouped by a map keyed on more than the dimension it walks is refused #1656 under the language's relation, and which shapes of it this package builds #1653. This is the bug of the four; those are gaps.
• No example or gallery page. A self-map is exercised by the new test module.
• feat(data): a relation may be keyed by several columns, so a generator's zone can change by period #1650 touches the same two functions. It is not a dependency either way — this branches off main so it can merge alone — but whichever lands second wants a small merge in _mapping and operator_at. I will do that when you merge one.

🤖 Generated with Claude Code

https://claude.ai/code/session_014pdMgEVkGBfh1f2AyiCSKA

---

Generated by Claude Code

[codspeed]

Merging this PR will not alter performance

✅ 24 untouched benchmarks
⏩ 58 skipped benchmarks¹

---

_{Comparing fix/self-map-walk-column (ff191bf) with main (1fd8926)²}

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. ↩

2. No successful run was found on main (28cceaf) during the generation of this report, so 1fd8926 was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩