interating over 1d array returns dimensionless numpy arrays in v3

[christine-e-smit]

Zarr version

3.1.2

Numcodecs version

0.16.1

Python Version

3.13.5

Operating System

Mac

Installation

using conda

Description

When I iterate over a 1D array, I get dimensionless numpy arrays out rather than just getting the data values. This behavior is different from zarr library version 2. In zarr library v2, I get the data values themselves. Is this new behavior intended?

Details of python setup where the script below works:
zarr version: 2.18.4
numcodecs version: 0.15.1
python version: 3.13.5

Steps to reproduce

# /// script
# requires-python = ">=3.11"
# dependencies = [
#   "zarr@git+https://github.com/zarr-developers/zarr-python.git@main",
#   "numpy"
# ]
# ///
#
# This script automatically imports the development branch of zarr to check for issues

import zarr
import numpy as np

z = zarr.zeros(shape=(2,), chunks=(2,), dtype=np.int32)
listed = list(z)
assert type(listed[0]) == np.int32

Additional output

No response

[d-v-b]

that's weird, and probably not intended. here's a related issue about changes in linear array iteration semantics in v3: #2529. I will try to figure out why list(array) is making a bunch of 0D arrays instead of a list of scalars.

[christine-e-smit]

Interesting. I had wondered how chunking interacted with iterating. I guess I figured that in the 1D case, it would be pretty straightforward to cache the chunks as the iterator moves along. But it sounds as though I should update our code right now to just grab the whole array at once and then iterate through it in memory since that will be much more efficient with the current implementation.

I had not thought at all about iterating over higher dimensional arrays in this fashion. When I found this issue, I did give iterating over a 2D array a try just because I was curious about what I'd get, but it feels me as though iterating through big, higher dimensional arrays is an anti-pattern. The 1D array I was iterating over was a dimension array and relatively small.

Thanks for the insight!