Reshaping Issue from 2025.12 onwards - awaiting new test data

[dwest77a]

[dwest77a]

Reshaping using single index (non-slice) should be possible for xarray DataArrays, but a bug with the reshaping prevents this. Unfortunately this is buried deep within the xarray model and has no current solution.

# Working solution
ds.var.isel(time=slice(0,1)).mean(dim='time')

# Does not work
ds.var.isel(time=0).compute()

[dwest77a]

Testing fixes in the issue40 branch, needs a new version of the ArrayPartition library with some modifications

[dwest77a]

New fixes cause an uncontrolled and unexplained segmentation fault seemingly from nowhere

[dwest77a]

Issue remains unresolved as of 16th December

[dwest77a]

Resolved by ditching previous attempts to fix using dropped_extent, now extent remains relative to the dimensions of the original data, while the shape is dependent on which parts of the extent are integer based. E.g for an extent [0, slice(0,1)] the shape is (1) as the first dimension will be dropped in the process of rendering this array partition

[dwest77a]

Just need to reset the test data so it uses local file paths, in order to work correctly on the github action

[dwest77a]

Also need to understand the effects on combining slices when an extent has dropped a dimension.

[dwest77a]

2026 Release 1 Actions:

• revisit combine slices function
• New test data relative paths
• Documentation for new changes
• Scaling and offset handling, consistent with cf-python approach