v1.4.1

v1.4.1 (2026-08-06)

Changed

• resample_time now requires uniform time steps by default when resampling. Irregular time coordinates that were previously resampled based on their mean timestep now raise a ValueError unless require_uniform_spacing=False is passed.
• Xarray-based drivers now reject parallel=True with lock=False when the active Dask scheduler is threaded, instead of forwarding an unsafe keyword-argument combination to xarray.open_mfdataset.
• Multi-file Dataset, GeoDataset and RasterDataset sources with {year} or {month} URI placeholders now check that all resolved periods are present in the opened data, including when local path separators differ between platforms.

Fixed

• SlippyTileDriver no longer crashes when an AWS credential provider raises while resolving (e.g. an expired SSO token); it now falls back to anonymous access, so public buckets remain reachable.
• readers.open_mfraster now uses the numeric basename as the concatenation index for files matched by a bare wildcard (e.g. *.tif matching 1.tif .. 12.tif) instead of the file-order fallback, and sorts the matched files so the concat order and variable name no longer depend on the filesystem glob order. (#1465)
• preprocessing.harmonise_dims no longer applies 0-360 to -180-180 longitude normalisation to projected datasets (x in metres, all > 180), which silently shifted the grid by -360. Normalisation is now only applied for geographic CRS; datasets without a CRS are left unchanged with a warning. (#1476)
• RasterDatasetAdapter now respects handle_nodata when requested variables are missing from the source: NoDataStrategy.WARN and NoDataStrategy.IGNORE return None (and warn) instead of always raising NoDataException. (#1407)
• A DataFrame source without an explicit driver now defaults to the pandas driver instead of incorrectly inferring geodataframe_table (which also claims .csv/.parquet) and failing validation. (#1403)