support mixed int/slice indexing

[dcherian]

There appears to be an upstream Xarray bug for mixed integer/slice indexing ds.isel(x=slice(10), y=1). The following is Claude's explanation:

rasterix/tests/test_indexing.py

Lines 26 to 44 in 0b84419

	def is_mixed_scalar_slice_indexer(indexers: dict[Hashable, int | slice]) -> bool:
	# TODO: Fix bug in RasterIndex with mixed scalar/slice indexing across dimensions
	# When you have scalar indexing on one dimension (e.g., y=0) and slice indexing
	# on another (e.g., x=slice(None, 1)), RasterIndex.isel() returns None for the
	# scalar dimension, dropping that index. This causes xarray to incorrectly handle
	# the coordinate variables - the sliced dimension's coordinate (x) maintains
	# dims ('x',) even though the data has been reduced by the scalar indexing.
	# This results in: "ValueError: dimensions ('x',) must have the same length as
	# the number of data dimensions, ndim=0"
	#
	# Example failing case: raster_da.isel(y=0, x=slice(None, 1))
	# - y=0 causes RasterIndex to return None for y dimension
	# - x=slice(None, 1) preserves RasterIndex for x dimension
	# - Result: coordinate variable x has wrong dimensionality
	#
	# For now, filter out these cases using hypothesis.assume()
	has_scalar = any(isinstance(v, int | np.integer) for v in indexers.values())
	has_slice = any(isinstance(v, slice) for v in indexers.values())
	return has_scalar and has_slice and len(indexers) > 1

cc @benbovy

[benbovy]

I've been looking into this.

The actual issue is indexing a coordinate variable that wraps a CoordinateTransform with a 1-length array or slice, i.e., here raster_da._coords["x"].isel(x=slice(0, 1)).

This is similar than #44 and will normally be fixed by pydata/xarray#10980.

[dcherian]

Looks like you're right.

[dcherian]

Looks like you were right!

[benbovy]

The GiHub mobile app was showing my previous comment in a very weird and useless way :-)