groupby().sum() returns wrong dims for a multi-dimensional DataArray grouper (fast path)

[FabianHofmann]

Likely a regression from #802

Note

created by AI

Bug

Variable.groupby(...).sum() returns wrong dimensions when grouping by a multi-dimensional DataArray grouper on its default (fast) path. The reduction happens over only one of the grouper's dimensions, and the other survives in the result — producing incorrect (and silently wrong) grouped constraints/expressions.

The use_fallback=True escape hatch (native xarray groupby) gives the correct result.

Root cause

In LinearExpressionGroupby.sum (linopy/expressions.py), the fast path does group = group.to_pandas() for a DataArray grouper. For a 2-D grouper this yields a DataFrame, which is then routed through the multikey encoding path — reducing over only the row axis and leaving the column axis in the output. A genuinely N-D grouper (whose values should define the groups over all its dims jointly) is not handled.

Minimal reproducer

import linopy, xarray as xr

# 2x2 grid, grouped into 2 groups by a 2-D grouper; extra dim 'k' should be preserved
grp = xr.DataArray([[1, 1], [2, 2]], dims=["i", "j"],
                   coords={"i": [0, 1], "j": [0, 1]}).rename("g")

m = linopy.Model()
v = m.add_variables(coords=[[0, 1], [0, 1], [0, 1]], dims=["i", "j", "k"], name="v")

print(v.groupby(grp).sum().dims)                    # ('j', 'k', 'g', '_term')  <- stray 'j', WRONG
print(v.groupby(grp).sum(use_fallback=True).dims)   # ('g', 'k', '_term')        <- correct

Plain xarray reduces the same 2-D grouper correctly over both dims, so the expected result is ('g', 'k', '_term') (group dim + the untouched k).

Impact

Silent incorrectness: expressions/constraints built via groupby().sum() with a multi-dim grouper are wrong unless the user knows to pass use_fallback=True. Surfaced while writing the examples/sudoku.ipynb example (grouping a (row, column, digit) variable by a 2-D (row, column) square-index grouper), which currently has to carry use_fallback=True as a workaround.

Environment

• linopy 0.8.0.post1.dev109+ge87efc1f4
• xarray 2026.4.0

Suggested fix

Detect a multi-dimensional DataArray grouper on the fast path and either handle it directly or route it to the native-xarray (fallback) implementation instead of to_pandas()-ing it into a multikey DataFrame.