bug: inserting a list into a two-dimensional array does not work as in numpy

[ombschervister]

Numpy supports inserting:

import numpy as np
a = np.array([(1, 2), (2, 3), (3, 4)])
a[0:1] = [5, 6]

Zarr gets an exception:

import zarr
za = zarr.array(a)
za[0:1] = [5, 6]

ValueError: parameter 'value': expected array with shape (1, 2), got (2,)

[ombschervister]

For arrays with structured types does not work at all.

Numpy supports inserting:

import numpy as np

t = [('c1', np.int), ('c2', np.float)]
a = np.array([(1, 2.0), (2, 3.0), (3, 4.0)], dtype=t)
a[0] = (4, 5.0)
a[0:1] = [(5, 6.0)]
a[0:2] = [(6, 7.0), (7, 8.0)]

Zarr gets an exception:

import zarr
za = zarr.array(a)
za[0] = (8, 9.0)
za[0:1] = [(5, 6.0)]   # ValueError: parameter 'value': expected array with shape (1,), got (1, 2)
za[0:2] = [(6, 7.0), (7, 8.0)]   # ValueError: parameter 'value': expected array with shape (2,), got (2, 2)

[martindurant]

How about

za[0:1] = [ [5, 6] ]

The upcasting of the shape (broadcasting) is a special case, perhaps it should be better handled indeed.