fix(array): reject setitem values that do not broadcast to the selection

[d-v-b]

🤖 AI text below 🤖

Writing a value whose shape does not broadcast to the selection now raises ValueError, as numpy does, instead of silently writing a truncated value; this also covers zarr.create(shape, data=...). Values that numpy broadcasts, such as a length-1 axis, a lower-rank value, or extra leading length-1 axes, are now written correctly for every selection type instead of raising or writing partial data.

Summary

import numpy as np, zarr
a = zarr.zeros((3,))
a[...] = np.arange(4)   # numpy raises; zarr wrote [0. 1. 2.] with no error

_set_selection passed the value to the codec pipeline unchecked, and the pipeline slices it with each chunk's out_selection. So a value longer than the selection was truncated, and values numpy accepts failed or wrote partial data depending on chunk layout, for example a (1,) value spanning several chunks, (3,) into (4, 3), a leading length-1 axis, or a (1,) value via vindex.

The value is now broadcast to the selection shape under numpy's assignment rules (_broadcast_to_selection) before any chunk is touched, raising numpy's could not broadcast input array from shape X into shape Y otherwise. 0-d values pass through unchanged so the pipeline's scalar fast paths still apply. Coordinate and mask selections broadcast to sel_shape and are then flattened, which replaces the ad-hoc flatten and element-count check in Array.set_coordinate_selection.

For reviewers

• A broadcast value reaches the codecs as a read-only, stride-0 view on the complete-chunk path. The test suite plus a manual check with v2 numcodecs compressors/filters (Blosc+Delta, Zlib, F order) and v3 Blosc found no codec that rejects it, but this is the part most worth a second look.
• The coordinate/mask error message changes from "Attempting to set a selection of N elements..." to numpy's wording. test_set_selection_rejects_value_with_wrong_rank was updated, and its docstring no longer describes an oindex write that chunked arrays accepted and sharded arrays rejected, since both now reject it.
• No existing issue was found for this.

Tests: one test covering the valid broadcasts against numpy (basic, integer axis, oindex, vindex, mask; chunked and sharded), one per error case (axis length mismatch; extra non-length-1 leading axis), plus the zarr.create(data=...) path. 37 of the 41 new cases fail on main; the 4 that pass are exact-shape controls.

Author attestation

• I am a human, these are my changes, and I have reviewed and understood every change and can explain why each is correct.

TODO

• Add unit tests and/or doctests in docstrings
• Add docstrings and API docs for any new/modified user-facing classes and functions
• New/modified features documented in docs/user-guide/*.md
• Changes documented as a new file in changes/
• GitHub Actions have all passed
• Test coverage is 100% (Codecov passes)

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 94.44%. Comparing base (9677081) to head (48a513c).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #4456      +/-   ##
==========================================
- Coverage   94.46%   94.44%   -0.02%     
==========================================
  Files          93       93              
  Lines       13234    13238       +4     
==========================================
+ Hits        12501    12503       +2     
- Misses        733      735       +2     

Files with missing lines	Coverage Δ
src/zarr/core/array.py	98.18% <100.00%> (+0.09%)	⬆️

... and 2 files with indirect coverage changes

🚀 New features to boost your workflow:

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