[BUG] crop_with_convex_hull fails when margin_thickness == 0.0

[arnavsharma990]

The Issue Description is Written with the help of Claude

Summary

When margin_thickness is set to 0.0, crop_with_convex_hull assigns the full tuple returned by create_convex_hull_mask(...) to da_mask, which later causes a type mismatch and runtime failure.

Location

mllam_data_prep/ops/cropping.py
Function: crop_with_convex_hull

Problem

In the zero-margin branch:

da_mask = create_convex_hull_mask(...)

However, create_convex_hull_mask returns:

(mask, hull_points)

Later in the same function:

_mask_with_common_dim(da_mask=da_mask, ds=ds)

_mask_with_common_dim expects da_mask to be an xarray.DataArray, but it receives a tuple instead.

This leads to a runtime error when the zero-margin path is triggered.

Why This Matters

Users may legitimately configure margin_thickness = 0.0 when they want cropping strictly along the convex hull boundary.

In this configuration the pipeline crashes, making this a correctness issue rather than just a minor edge case.

Suggested Fix

Unpack the return value explicitly in the zero-margin branch:

da_mask, _ = create_convex_hull_mask(...)

This keeps the expected DataArray type flowing into _mask_with_common_dim.

Possible Follow-up

It may also be helpful to add a regression test that runs the convex hull cropping with:

margin_thickness = 0.0

to ensure the code path remains stable.

Expected Impact

Small patch with high correctness impact. This change should not affect existing behavior except fixing the failure when zero margin is used.

[arnavsharma990]

I'd be happy to work on a fix for this if it sounds reasonable

[RajdeepKushwaha5]

Thanks @arnavsharma990 for filing this — the core diagnosis is correct. A few additional details I noticed while tracing through the code that may be worth capturing in the fix:

1. Exact error: The crash is specifically AttributeError: 'tuple' object has no attribute 'dims' at line 365 inside _mask_with_common_dim(), when it tries to access da_mask.dims.

2. Docstring / type hint mismatch on create_convex_hull_mask: The function signature says -> xr.DataArray (line 33), but it actually returns (da_interior_mask, chull_lat_lons) — a Tuple[xr.DataArray, xr.Dataset]. The docstring does document both return values, but the type annotation is wrong. This should probably be corrected to -> Tuple[xr.DataArray, xr.Dataset] as part of the fix to prevent future confusion.

3. return_mask=True path is also affected: When margin_thickness == 0.0 and return_mask=True, line 367-368 returns (ds_cropped, da_mask) where da_mask is still the unbroken tuple. So callers would receive (Dataset, (DataArray, Dataset)) instead of the expected (Dataset, DataArray). The unpack fix da_mask, _ = ... would resolve this as well.

4. Regression test should be mandatory, not optional: The existing test_convex_hull_cropping.py never exercises margin_thickness=0.0. A test parametrized with margin_thickness=0.0 (and optionally return_mask=True) should be included in the PR, not treated as a follow-up.

I'm planning to work on a PR for this if nobody else is already on it.