Change in indexing behaviour with numpy ints

[benjeffery]

After 7c31f04 there was a breaking change to indexing behavior when a numpy integer type is passed:

For the following array:

z = zarr.group()
a = z.create_dataset("test",shape=(10,))
a[:]=range(10)

Before this commit:
a[np.int64(1)] has the value 1.0
After:
a[np.int64(1)] has the value [1.]

For both cases a[1] stays as 1.0 as expected. Was this change expected? This happens a lot when iterating over one numpy array and indexing with it as the iterable is of the numpy type. The fix in our application was to cast to a python int before indexing.

EDIT: Also note that numpy indexing returns the single value as was the case before this commit.

[raphaelquast]

I was just about to write the exact same issue... (stumbled upon it while trying to read an encoded dict)

root = zarr.group()
root.create("asdf", shape=(10,), dtype=object, object_codec=numcodecs.JSON())
root.asdf[0] = dict(a=1, b=2)

# using a normal python int to get the data
root.asdf[0]
#>>> {'a': 1, 'b': 2}

# using a numpy int to get the data
ids = np.array([0, 1, 2])
root.asdf[ids[0]]
#>>> array([{'a': 1, 'b': 2}], dtype=object)

[sanketverma1704]

This issue has been fixed in patch release 2.11.1. 🎉

[benjeffery]

This issue has been fixed in patch release 2.11.1. tada

Woop! Thankyou for the quick release!