`move("/", "/dataset")` makes data unreachable via `list_dir` from root

[TomNicholas]

Description

After moving the root group to a subpath and committing, list_dir("/") returns nothing even though the data is still present in storage (visible via list()). Opening the store with zarr/xarray from the root would see an empty store.

Reproducer

import asyncio
import tempfile
import icechunk as ic
import xarray as xr

async def main():
    air_temp = xr.tutorial.open_dataset("air_temperature")

    storage = ic.local_filesystem_storage(tempfile.TemporaryDirectory().name)
    repo = ic.Repository.create(storage)

    session = repo.writable_session("main")
    air_temp.to_zarr(session.store)
    session.commit("wrote tutorial dataset to root group")

    # Before the move — list_dir("/") works as expected
    session = repo.readonly_session(branch="main")
    print('--- before move, list_dir("/") ---')
    async for r in session.store.list_dir("/"):
        print(r)

    # Move the root group to /dataset
    session = repo.rearrange_session("main")
    session.move("/", "/dataset")
    session.commit("moved root to /dataset")

    session = repo.readonly_session(branch="main")

    print('--- after move, list_dir("/") ---')
    async for r in session.store.list_dir("/"):
        print(r)

    print('--- after move, list_dir("/dataset") ---')
    async for r in session.store.list_dir("/dataset"):
        print(r)

    print('--- after move, list_dir("dataset") ---')
    async for r in session.store.list_dir("dataset"):
        print(r)

    print("--- after move, list() ---")
    async for r in session.store.list():
        print(r)

asyncio.run(main())

Output

--- before move, list_dir("/") ---
zarr.json
air
lat
lon
time
--- after move, list_dir("/") ---
--- after move, list_dir("/dataset") ---
--- after move, list_dir("dataset") ---
zarr.json
air
lat
lon
time
--- after move, list() ---
dataset/zarr.json
dataset/air/zarr.json
dataset/lat/zarr.json
dataset/lon/zarr.json
dataset/time/zarr.json
dataset/air/c/0/0/0
dataset/air/c/0/0/1
dataset/air/c/0/1/0
dataset/air/c/0/1/1
dataset/air/c/1/0/0
dataset/air/c/1/0/1
dataset/air/c/1/1/0
dataset/air/c/1/1/1
dataset/air/c/2/0/0
dataset/air/c/2/0/1
dataset/air/c/2/1/0
dataset/air/c/2/1/1
dataset/air/c/3/0/0
dataset/air/c/3/0/1
dataset/air/c/3/1/0
dataset/air/c/3/1/1
dataset/lat/c/0
dataset/lon/c/0
dataset/time/c/0

Expected

list_dir("/") returns dataset as a child. list_dir("/dataset") and list_dir("dataset") behave identically.

Actual

list_dir("/") and list_dir("/dataset") return nothing. Only list_dir("dataset") (no leading slash) returns the group contents.

Notes

Possibly two bugs:

1. move("/", ...) leaves no root group, so list_dir("/") finds no children even though storage keys exist under dataset/.
2. Leading-slash path normalization in list_dir — /dataset vs dataset should be equivalent.

[ianhi]

session.move("/", "/dataset") absoultely should not be allowed

[ianhi]

ok on further discussion this is a reasonable thing for a user to want to do. i.e. moving from a top level zarr hierarchy to a more organized one. This is close to moving all arrays under the root group to be under the new group, but not exactly because the root group may also have metadata that needs to be moved.

I really thought we disallowed this become your own child behavior and that the root was slipping through, but apparently it is also allowed any group

import icechunk as ic
import zarr
from zarr.core.group import GroupMetadata


storage = ic.storage.in_memory_storage()
repo = ic.Repository.create(storage)

session = repo.writable_session("main")
root = zarr.open(session.store)
list(root.create_hierarchy({'a/b/c': GroupMetadata()}))

print("before move")
print(root.tree())

session.commit('msg')
session = repo.rearrange_session('main')
session.move('/a', '/a/c')
root = zarr.open(session.store)
print("after move")
root.tree()

before move
/
└── a
    └── b
        └── c

after move

/