Allow concatenation of string coordinates of differing width (#6676)

[gaoflow]

🚀 Pull Request

Description

Closes #6676.

Two cubes whose string coordinate differs only in width (i.e. dtype, such as <U1 vs <U5) could not be concatenated:

cube_a = iris.cube.Cube(
    [0, 1], long_name="test",
    dim_coords_and_dims=[(iris.coords.DimCoord([0, 1], long_name="dim"), 0)],
    aux_coords_and_dims=[(iris.coords.AuxCoord(["1", "2"], long_name="example"), 0)],
)
cube_b = iris.cube.Cube(
    [10, 12, 13], long_name="test",
    dim_coords_and_dims=[(iris.coords.DimCoord([10, 11, 12], long_name="dim"), 0)],
    aux_coords_and_dims=[(iris.coords.AuxCoord(["1", "123", "12345"], long_name="example"), 0)],
)
iris.cube.CubeList([cube_a, cube_b]).concatenate_cube()

ConcatenateError: failed to concatenate into a single cube.
  Auxiliary coordinates metadata differ: example != example

Cause

When building the coordinate signature, _CoordMetaData records each coordinate's points_dtype/bounds_dtype, and __eq__ compares them exactly. For string coordinates, <U1 and <U5 are different dtypes, so the metadata was reported as differing even though the coordinates are otherwise compatible.

Fix

When comparing coordinate signatures, collapse string dtypes (kind "U"/"S") to their kind, so that differing widths no longer block concatenation. numpy promotes the points to a common width when they are joined, so the result coordinate gets the wider dtype (<U5 above). Genuine dtype-kind differences (e.g. string vs integer) are unaffected and still rejected.

Verification

• New TestStringAuxCoordWidths:
  • test_different_widths — the two cubes above now concatenate; the result coordinate has dtype <U5 and the expected points. Fails on main, passes here.
  • test_different_dtype_kind_still_rejected — string-vs-integer coordinates still raise ConcatenateError.
• Full tests/unit/concatenate/, tests/integration/concatenate/ and tests/test_concatenate.py suites pass (283 + 54 tests).
• ruff check / ruff format clean.

[codecov]

Codecov Report

❌ Patch coverage is 50.00000% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 90.15%. Comparing base (746f0a6) to head (6591a46).

Files with missing lines	Patch %	Lines
lib/iris/_concatenate.py	50.00%	1 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #7125      +/-   ##
==========================================
- Coverage   90.15%   90.15%   -0.01%     
==========================================
  Files          91       91              
  Lines       24985    24989       +4     
  Branches     4685     4687       +2     
==========================================
+ Hits        22526    22528       +2     
- Misses       1682     1683       +1     
- Partials      777      778       +1     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[trexfeathers]

Thanks @gaoflow! A few comments for you this time

[gaoflow]

Updated in 6591a468 and rebased onto current main.

Review follow-up:

• moved string-width normalization into _CoordMetaData.__new__, so the metadata stores the normalized dtype consistently instead of special-casing __eq__
• shortened the implementation comment
• reused the test helper for the integer-coordinate case
• matched the existing (result,) = concatenate(..., True) style
• replaced the hand-written latest.rst entry with changelog/7125.bugfix.rst

Verification after the rebase:

uv run --with pytest --with pytest-mock --with requests --with filelock --with-editable . python -m pytest lib/iris/tests/unit/concatenate/test_concatenate.py::TestStringAuxCoordWidths -q
uv run --with pytest --with pytest-mock --with requests --with filelock --with-editable . python -m pytest lib/iris/tests/unit/concatenate/test_concatenate.py -q
uv run --with towncrier towncrier build --draft --version 0.0
ruff check lib/iris/_concatenate.py
ruff format --check lib/iris/_concatenate.py lib/iris/tests/unit/concatenate/test_concatenate.py
git diff --check origin/main...HEAD

The full touched test file passed with 43 passed.

[trexfeathers]

Another valuable contribution, thanks @gaoflow!