restore v2 behavior

src/zarr/codecs/pipeline.py

@@ -17,7 +17,7 @@
 from zarr.core.common import ChunkCoords, concurrent_map
 from zarr.core.config import config
 from zarr.core.indexing import SelectorTuple, is_scalar, is_total_slice
-from zarr.core.metadata.v2 import default_fill_value
+from zarr.core.metadata.v2 import _default_fill_value
 from zarr.registry import register_pipeline
 
 if TYPE_CHECKING:
@@ -256,7 +256,7 @@ async def read_batch(
                         # validated when decoding the metadata, but we support reading
                         # Zarr V2 data and need to support the case where fill_value
                         # is None.
-                        fill_value = default_fill_value(dtype=chunk_spec.dtype)
+                        fill_value = _default_fill_value(dtype=chunk_spec.dtype)
 
                     out[out_selection] = fill_value
         else:
@@ -287,7 +287,7 @@ async def read_batch(
                 else:
                     fill_value = chunk_spec.fill_value
                     if fill_value is None:
-                        fill_value = default_fill_value(dtype=chunk_spec.dtype)
+                        fill_value = _default_fill_value(dtype=chunk_spec.dtype)
                     out[out_selection] = fill_value
 
     def _merge_chunk_array(

src/zarr/core/metadata/v2.py

@@ -302,7 +302,7 @@ def parse_fill_value(fill_value: object, dtype: np.dtype[Any]) -> Any:
     return fill_value
 
 
-def default_fill_value(dtype: np.dtype[Any]) -> Any:
+def _default_fill_value(dtype: np.dtype[Any]) -> Any:
     """
     Get the default fill value for a type.
 
@@ -315,4 +315,9 @@ def default_fill_value(dtype: np.dtype[Any]) -> Any:
     This is useful for reading Zarr V2 arrays, which allow the fill
     value to be unspecified.
     """
-    return dtype.type(0)
+    if dtype.kind == "S":
+        return b""
+    elif dtype.kind == "U":
+        return ""
+    else:
+        return dtype.type(0)

tests/v3/test_v2.py

@@ -1,5 +1,6 @@
 import json
 from collections.abc import Iterator
+from typing import Any
 
 import numpy as np
 import pytest
@@ -35,12 +36,24 @@ def test_simple(store: StorePath) -> None:
     assert np.array_equal(data, a[:, :])
 
 
-def test_implicit_fill_value(store: StorePath) -> None:
-    arr = zarr.open_array(store=store, shape=(4,), fill_value=None, zarr_format=2)
+@pytest.mark.parametrize(
+    ("dtype", "fill_value"),
+    [
+        ("bool", False),
+        ("int64", 0),
+        ("float64", 0.0),
+        ("|S1", b""),
+        ("|U1", ""),
+        ("object", 0),
+        (str, ""),
+    ],
+)
+def test_implicit_fill_value(store: StorePath, dtype: str, fill_value: Any) -> None:
+    arr = zarr.open_array(store=store, shape=(4,), fill_value=None, zarr_format=2, dtype=dtype)
     assert arr.metadata.fill_value is None
     assert arr.metadata.to_dict()["fill_value"] is None
     result = arr[:]
-    expected = np.zeros(arr.shape, dtype=arr.dtype)
+    expected = np.full(arr.shape, fill_value, dtype=dtype)
     np.testing.assert_array_equal(result, expected)