zarr.open_array duplicates `path` key?

[TomAugspurger]

Zarr version

v3

Numcodecs version

n/a

Python Version

n/a

Operating System

n/a

Installation

n/a

Description

I'm looking into some of the xarray tests, and noticed this strange(?) change from V2. It seems like zarr.open_array with path="a" will write the data to a/a/, so that the metadata is at key a/a/zarr.json. I would have expected it to be at a/zarr.json. Is this expected?

Steps to reproduce

# v3
import zarr
import shutil
shutil.rmtree("/tmp/data.zarr", ignore_errors=True)

store = zarr.store.LocalStore(root="/tmp/data.zarr", mode="w")
a = zarr.open_array(store=store, path="a", shape=(4,))
b = zarr.open_array(store=store, path="b", shape=(4,))
[x async for x in store.list()]

gives

['a/a/zarr.json', 'b/b/zarr.json']

and the arrays must be opened with zarr.open_array(store=store, path="a/a").

Contrast that with V2, where you have

import zarr


store = "/tmp/v2.zarr"
a = zarr.open_array(store=store, path="a", shape=(4,))
b = zarr.open_array(store=store, path="b", shape=(4,))

g = zarr.open_group("/tmp/v2.zarr")
list(g.array_keys())

give ['a', 'b'] and you open the arrays with zarr.open_array(store=store, path="a").

Additional output

No response

[dcherian]

I concur that there is something bizarre here. I struggled with it:

zarr-python/src/zarr/testing/strategies.py

Lines 80 to 97 in ac6c6a3

	# TODO: clean this up
	# if path is None and name is None:
	# array_path = None
	# array_name = None
	# elif path is None and name is not None:
	# array_path = f"{name}"
	# array_name = f"/{name}"
	# elif path is not None and name is None:
	# array_path = path
	# array_name = None
	# elif path == "/":
	# assert name is not None
	# array_path = name
	# array_name = "/" + name
	# else:
	# assert name is not None
	# array_path = f"{path}/{name}"
	# array_name = "/" + array_path

Would be good to uncomment (and simplify) once someone figures this out :)

[TomAugspurger]

Ah, looks like if path is provided it's appended to store path at

zarr-python/src/zarr/api/asynchronous.py

Lines 889 to 890 in ac6c6a3

	if path is not None:
	store_path = store_path / path

Then when we go to create the array in create we pass it again

zarr-python/src/zarr/api/asynchronous.py

Line 900 in ac6c6a3

path=path,

, despite it already being present in the store_path we provide as store.

I'll see if anything breaks when removing that.

[jhamman]

Nice catch @TomAugspurger!