fix: name offending elements when writing object arrays with non-strings to h5ad

[SajalDevX]

• Closes #
• Tests added

What

Writing an object array that holds non-str items (None, nan, ints, …) as a string array to h5ad currently fails with h5py's bare

TypeError: Can't implicitly convert non-string objects to strings
Error raised while writing key 'k' of <class 'h5py._hl.group.Group'> to /uns

The key is reported (thanks to #1272), but not what inside the array is wrong, and the #726 thread shows how often people hit this without knowing which values to fix. With this PR the same case raises

TypeError: Cannot write object array as string array: 1 of 2 elements are not str (found types: NoneType), e.g. element (1,) is None. Convert the array to str before writing. For missing values, use a categorical or a nullable pandas string dtype instead of None/nan.
Error raised while writing key 'k' of <class 'h5py._hl.group.Group'> to /uns

chained (from) to h5py's original error.

How

write_vlen_string_array (h5py) wraps the create_dataset call in a try/except TypeError; on failure a small helper scans the object array for non-str items and raises the descriptive error. If the array is not an object array, or all items are str, the original exception is re-raised untouched. Nothing runs on the happy path, so there is no cost for valid arrays.

Reproducer:

import numpy as np, anndata as ad
a = ad.AnnData(X=np.zeros((2, 2), dtype=np.float32))
a.uns["k"] = np.array(["a", None], dtype=object)
a.write_h5ad("x.h5ad")

Note on zarr

write_vlen_string_array_zarr behaves differently: it silently coerces these items to strings (np.array(["a", 1], dtype=object) round-trips as ["a", "1"]). I left that alone and only skip the new test for zarr, but it may be worth making the two backends consistent — happy to follow up either way if you tell me which direction you prefer.

[codecov]

Codecov Report

❌ Patch coverage is 82.35294% with 3 lines in your changes missing coverage. Please review.
✅ Project coverage is 85.78%. Comparing base (c2c1c37) to head (984d1ac).
✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
src/anndata/_io/specs/methods.py	82.35%	3 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2667      +/-   ##
==========================================
- Coverage   87.62%   85.78%   -1.85%     
==========================================
  Files          49       49              
  Lines        8129     8145      +16     
==========================================
- Hits         7123     6987     -136     
- Misses       1006     1158     +152     

Files with missing lines	Coverage Δ
src/anndata/_io/specs/methods.py	91.38% <82.35%> (-0.61%)	⬇️

... and 7 files with indirect coverage changes