Repository navigation
ux.open_dataset currently supports structured data #1854
Description
Version
v2026.10.0
How did you install UXarray?
Source
What happened?
ux.open_dataset() currently supports structured data (e.g. a lat/lon grid), silently returning a reasonable-looking answer. However, that is actually not intentional. For more details see: #1827 (comment).
What did you expect to happen?
Because opening structured data with ux.open_dataset() is not intended to be supported and is not currently being tested, uxarray should avoid silently returning a reasonable-looking answer which might actually be wrong. So, I would expect it to crash, or at least raise a warning, when pointed at structured data.
The crash or warning message should maybe point out that the UxDataset.from_structured() or UxDataset.from_xarray() could be considered instead, as the standard entry points for loading structured data into uxarray.
Can you provide a MCVE to repoduce the bug?
import numpy as np
import xarray as xr
import uxarray as ux
ds = xr.Dataset(
{
"t": (("time", "lat", "lon"), np.random.rand(2, 4, 8)),
"s": (("lat", "lon"), np.random.rand(4, 8)),
"z": (("time",), np.ones(2)),
},
coords={
"lat": ("lat", np.linspace(-80, 80, 4), {"standard_name": "latitude"}),
"lon": ("lon", np.linspace(0, 350, 8), {"standard_name": "longitude"}),
"time": [0, 1],
},
)
ux.open_dataset(ds, ds) # grid, data
# silently returns a UxDataset with sizes {'time': 2, 'n_face': 32};
# I expected a crash or at least a warning, instead.Metadata
Metadata
Assignees
Labels
Type
Fields
Priority
Projects
- Status
Activity
erogluorhan commentedon Oct 27, 2026
I wouldn't call this a bug; however, I believe that we should still avoid this to guide structured grid access through specialized API (
from_structured()) only (I agree withux.open_dataset()not being intended to support structured grids).I'd like to share what I was able to track about how this might have happened so far:
To address the #725, #1043, and #1044 issues, PR #1040 was merged.
Key detail: This PR didn't intend to open doors for
open_dataset()to support structured grids (i.e. the only feature that was desired to be added wasfrom_structured()for our objects. However, we didn't test callingopen_dataset()directly on a structured grid/data either.Since this PR introduced
_is_structured()as a fallback inside_parse_grid_type(), and it is shared across multiple APIs, and_map_dims_to_ugrid()has a "structured" branch, any CF-compliant structured dataset that doesn't match another grid spec gets silently, unintentionally, auto-detected and converted — throughopen_dataset()/open_grid()too.As emphasized, this was unintended and needs to be avoided because if the user accidentally uses it this way and sees it working for that case, it becomes misleading.
Sevans711 commentedon Nov 3, 2026
Adding a quick note about something Orhan noticed during review of #1827. The new
_map_dataset_to_ugrid(ds, grid)introduced these lines for supporting structured data:These lines weren't really necessary, but they also don't affect any tests using unstructured data, so that PR wasn't blocked on it. These lines should most likely be removed as part of fixing this issue / tightening
ux.open_dataset()to no longer support structured data.