Incorrect shape computation with getitem and structured numpy array

[louisletoumelin]

Dear all,

When trying to access an element from a numpy structured array in dask, the dask array shape is not correctly estimated.

On my real world use case, I try to read lazily from a binary file using np.frombuffer/dask and then save my variables in a xarray dataset.

Thanks,

import dask.array as da
import numpy as np

buffer = da.random.randint(10,size=(100,4), dtype=np.uint8)
structured_array = da.map_blocks(np.frombuffer, buffer, ">H, >B, >B", dtype=">H, >B, >B")
print(structured_array["f1"].shape)  # print (100, 4), not ok
print(structured_array["f1"].compute().shape)  # print (100,), ok

Environment:

• Dask version: '2023.11.0'
• Python version: 3.12.2
• Operating System: Ubuntu 22.04.4 LTS
• Install method (conda, pip, source): conda