Remove ZARR_DEFAULT_FILL_VALUE lookup

conftest.py

@@ -10,6 +10,7 @@
 import numpy as np
 import pytest
 import xarray as xr
+import zarr
 from obstore.store import LocalStore
 from xarray.core.variable import Variable
 
@@ -72,10 +73,14 @@ def local_registry():
     return ObjectStoreRegistry({"file://": LocalStore()})
 
 
-@pytest.fixture()
-def zarr_store_scalar(tmpdir):
-    import zarr
+@pytest.fixture(params=["int8", "uint8", "float32"])
+def zarr_array_fill_value(request):
+    store = zarr.storage.MemoryStore()
+    return zarr.create_array(store=store, shape=(), dtype=request.param)
 
+
+@pytest.fixture()
+def zarr_store_scalar():
     store = zarr.storage.MemoryStore()
     zarr_store_scalar = zarr.create_array(store=store, shape=(), dtype="int8")
     zarr_store_scalar[()] = 42

docs/releases.md

@@ -8,6 +8,10 @@
 
 ### Bug fixes
 
+- `ZarrParser` no longer uses `ZARR_DEFAULT_FILL_VALUE` lookup to infer missing `fill_value`.
+  ([#666](https://github.com/zarr-developers/VirtualiZarr/pull/812)).
+  By [Raphael Hagen](https://github.com/norlandrhagen).
+
 ### Documentation
 
 ### Internal changes

virtualizarr/parsers/zarr.py

@@ -5,7 +5,7 @@
 from pathlib import Path  # noqa
 from typing import TYPE_CHECKING, Any, Hashable
 
-import numpy as np
+import zarr
 from zarr.api.asynchronous import open_group as open_group_async
 from zarr.core.metadata import ArrayV3Metadata
 from zarr.storage import ObjectStore
@@ -20,23 +20,13 @@
 from virtualizarr.registry import ObjectStoreRegistry
 from virtualizarr.vendor.zarr.core.common import _concurrent_map
 
-FillValueT = bool | str | float | int | list | None
-
-ZARR_DEFAULT_FILL_VALUE: dict[str, FillValueT] = {
-    # numpy dtypes's hierarchy lets us avoid checking for all the widths
-    # https://numpy.org/doc/stable/reference/arrays.scalars.html
-    np.dtype("bool").kind: False,
-    np.dtype("int").kind: 0,
-    np.dtype("float").kind: 0.0,
-    np.dtype("complex").kind: [0.0, 0.0],
-    np.dtype("datetime64").kind: 0,
-}
-
 if TYPE_CHECKING:
     import zarr
 
+ZarrArrayType = zarr.AsyncArray | zarr.Array
+
 
-async def get_chunk_mapping_prefix(zarr_array: zarr.AsyncArray, path: str) -> dict:
+async def get_chunk_mapping_prefix(zarr_array: ZarrArrayType, path: str) -> dict:
     """Create a dictionary to pass into ChunkManifest __init__"""
 
     # TODO: For when we want to support reading V2 we should parse the /c/ and "/" between chunks
@@ -70,26 +60,21 @@ async def get_chunk_mapping_prefix(zarr_array: zarr.AsyncArray, path: str) -> di
     }
 
 
-async def build_chunk_manifest(zarr_array: zarr.AsyncArray, path: str) -> ChunkManifest:
+async def build_chunk_manifest(zarr_array: ZarrArrayType, path: str) -> ChunkManifest:
     """Build a ChunkManifest from a dictionary"""
     chunk_map = await get_chunk_mapping_prefix(zarr_array=zarr_array, path=path)
     return ChunkManifest(chunk_map)
 
 
-def get_metadata(zarr_array: zarr.AsyncArray[Any]) -> ArrayV3Metadata:
-    fill_value = zarr_array.metadata.fill_value
-    if fill_value is not None:
-        fill_value = ZARR_DEFAULT_FILL_VALUE[zarr_array.metadata.fill_value.dtype.kind]
-
+def get_metadata(zarr_array: ZarrArrayType) -> ArrayV3Metadata:
     zarr_format = zarr_array.metadata.zarr_format
-
     if zarr_format == 2:
         # TODO: Once we want to support V2, we will have to deconstruct the
         # zarr_array codecs etc. and reconstruct them with create_v3_array_metadata
         raise NotImplementedError("Reading Zarr V2 currently not supported.")
 
     elif zarr_format == 3:
-        return zarr_array.metadata
+        return zarr_array.metadata  # type: ignore[return-value]
 
     else:
         raise NotImplementedError("Zarr format is not recognized as v2 or v3.")

virtualizarr/tests/test_parsers/test_zarr.py

@@ -1,13 +1,16 @@
 import numpy as np
 import pytest
+import zarr
 from obstore.store import LocalStore
 
 from virtualizarr import open_virtual_dataset
 from virtualizarr.manifests import ManifestArray
 from virtualizarr.parsers import ZarrParser
-from virtualizarr.parsers.zarr import get_chunk_mapping_prefix
+from virtualizarr.parsers.zarr import get_chunk_mapping_prefix, get_metadata
 from virtualizarr.registry import ObjectStoreRegistry
 
+ZarrArrayType = zarr.AsyncArray | zarr.Array
+
 
 @pytest.mark.parametrize(
     "zarr_store",
@@ -105,7 +108,7 @@ def test_virtual_dataset_zarr_attrs(self, zarr_store):
                 assert expected == actual
 
 
-def test_scalar_get_chunk_mapping_prefix(zarr_store_scalar):
+def test_scalar_get_chunk_mapping_prefix(zarr_store_scalar: ZarrArrayType):
     # Use a scalar zarr store with a /c/ representing the scalar:
     # https://zarr-specs.readthedocs.io/en/latest/v3/chunk-key-encodings/default/index.html#description
 
@@ -118,3 +121,9 @@ def test_scalar_get_chunk_mapping_prefix(zarr_store_scalar):
     )
     assert chunk_map["c"]["offset"] == 0
     assert chunk_map["c"]["length"] == 10
+
+
+def test_get_metadata(zarr_array_fill_value: ZarrArrayType):
+    # Check that the `get_metadata` function is assigning fill_values
+    zarr_array_metadata = get_metadata(zarr_array=zarr_array_fill_value)
+    assert zarr_array_metadata.fill_value == zarr_array_fill_value.metadata.fill_value