is dot syntax gone now?

[christine-e-smit]

Zarr version

v3.0.6

Numcodecs version

0.15.1

Python Version

3.13.3

Operating System

M<ac

Installation

using conda

Description

I don't actually know what you call this, but I'm calling it dot syntax for accessing arrays. With zarr version 2.18.4, I can do the following:

import zarr
import numpy as np
z = zarr.open("two.zarr",mode="w")
d = z.create_dataset("d",shape=(2),dtype=np.int32)
# this works!
z.d

But with zarr version 3.0.6, the following throws an exception:

import zarr
import numpy as np
z = zarr.open("three.zarr",mode="w")
d = z.create_array("d",shape=(2),dtype=np.int32)
# this now throws the exception "AttributeError: 'Group' object has no attribute 'd'"
z.d

Is this expected? Unexpected? I didn't see anything in the 3.0 Migration Guide, but it wouldn't surprise me if I missed something.

Steps to reproduce

import zarr
import numpy as np
z = zarr.open("three.zarr",mode="w")
d = z.create_array("d",shape=(2),dtype=np.int32)
# this now throws the exception "AttributeError: 'Group' object has no attribute 'd'"
z.d

Additional output

No response

[d-v-b]

hi @christine-e-smit, yes the "." syntax for getting group members has been removed in 3.0. To get a member of a group named "foo", use group["foo"]

I suspect we simply failed to mention this in the migration guide. I can fix this shortly.

[christine-e-smit]

Thank you for the clarification. I've submitted a pull request with some suggested language for the documentation.

[rabernat]

@christine-e-smit - did you find this syntax valuable?

For context, IIRC it was removed because many IDEs (e.g. JupyterLab, VCCode) will introspect python objects to provide auto-completion. With this dot syntax, that would actually trigger listing of groups, reading metadata, etc. operations with significant side effects and performance implications.

[christine-e-smit]

Interesting! I never personally ran into the IDE issue in Visual Studio Code, but I can see how it could be painful. I think I've used both g.x and g["x"] syntax without thinking terribly deeply about it, so I don't think I'll miss the "." syntax. I'm more concerned that finding all the instances where we used "." syntax is going to be tricky because it's not something you can just grep for. But if it's causing that many issues with IDEs, then I understand why you've dropped it.