refactor: use backported sentinel type

changes/4046.misc.md

@@ -0,0 +1 @@
+Replace the internal `DefaultFillValue` sentinel class with a `typing_extensions.Sentinel`, and raise the minimum `typing_extensions` version to 4.14.

pyproject.toml

@@ -35,7 +35,7 @@ dependencies = [
     'numpy>=2',
     'numcodecs>=0.14',
     'google-crc32c>=1.5',
-    'typing_extensions>=4.13',
+    'typing_extensions>=4.14',
     'donfig>=0.8',
 ]
 
@@ -256,7 +256,7 @@ extra-dependencies = [
     'fsspec==2023.10.0',
     's3fs==2023.10.0',
     'universal_pathlib==0.2.0',
-    'typing_extensions==4.13.*',
+    'typing_extensions==4.14.*',
     'donfig==0.8.*',
     'obstore==0.5.*',
 ]

src/zarr/core/array.py

@@ -18,7 +18,7 @@
 from warnings import warn
 
 import numpy as np
-from typing_extensions import deprecated
+from typing_extensions import Sentinel, deprecated
 
 import zarr
 from zarr.abc.codec import ArrayArrayCodec, ArrayBytesCodec, BytesBytesCodec, Codec
@@ -162,30 +162,26 @@
 # Array and AsyncArray are defined in the base ``zarr`` namespace
 __all__ = [
     "DEFAULT_FILL_VALUE",
-    "DefaultFillValue",
     "create_codec_pipeline",
     "parse_array_metadata",
 ]
 
 logger = getLogger(__name__)
 
 
-class DefaultFillValue:
-    """
-    Sentinel class to indicate that the default fill value should be used.
-
-    This class exists because conventional values used to convey "defaultness" like ``None`` or
-    ``"auto"` are ambiguous when specifying the fill value parameter of a Zarr array.
-    The value ``None`` is ambiguous because it is a valid fill value for Zarr V2
-    (resulting in ``"fill_value": null`` in array metadata).
-    A string like ``"auto"`` is ambiguous because such a string is a valid fill value for an array
-    with a string data type.
-    An instance of this class lies outside the space of valid fill values, which means it can
-    unambiguously express that the default fill value should be used.
-    """
-
+DEFAULT_FILL_VALUE = Sentinel("DEFAULT_FILL_VALUE")
+"""
+Sentinel indicating that the default fill value should be used.
 
-DEFAULT_FILL_VALUE = DefaultFillValue()
+This sentinel exists because conventional values used to convey "defaultness" like `None` or
+`"auto"` are ambiguous when specifying the fill value parameter of a Zarr array.
+The value `None` is ambiguous because it is a valid fill value for Zarr V2
+(resulting in `"fill_value": null` in array metadata).
+A string like `"auto"` is ambiguous because such a string is a valid fill value for an array
+with a string data type.
+This sentinel lies outside the space of valid fill values, which means it can
+unambiguously express that the default fill value should be used.
+"""
 
 
 def _chunk_sizes_from_shape(
@@ -541,9 +537,9 @@ def _create_metadata_v3(
         else:
             chunk_key_encoding_parsed = chunk_key_encoding
 
-        if isinstance(fill_value, DefaultFillValue) or fill_value is None:
-            # Use dtype's default scalar for DefaultFillValue sentinel
-            # For v3, None is converted to DefaultFillValue behavior
+        if fill_value is DEFAULT_FILL_VALUE or fill_value is None:
+            # Use dtype's default scalar for the DEFAULT_FILL_VALUE sentinel
+            # For v3, None is converted to DEFAULT_FILL_VALUE behavior
             fill_value_parsed = dtype.default_scalar()
         else:
             fill_value_parsed = fill_value
@@ -625,8 +621,8 @@ def _create_metadata_v2(
         if dimension_separator is None:
             dimension_separator = "."
 
-        # Handle DefaultFillValue sentinel
-        if isinstance(fill_value, DefaultFillValue):
+        # Handle the DEFAULT_FILL_VALUE sentinel
+        if fill_value is DEFAULT_FILL_VALUE:
             fill_value_parsed: Any = dtype.default_scalar()
         else:
             # For v2, preserve None as-is (backward compatibility)