fix: raise a validation error for out-of-bounds integer fill values

changes/4492.bugfix.md

@@ -0,0 +1 @@
+Raise a `ValueError` for out-of-bounds integer fill values instead of leaking NumPy's `OverflowError`, both when reading stored metadata and when casting values at array creation.

src/zarr/core/dtype/npy/int.py

@@ -178,6 +178,26 @@ def _cast_scalar_unchecked(self, data: IntLike) -> Scalar:
 
         return self.to_native_dtype().type(data)  # type: ignore[return-value]
 
+    def _check_int_bounds(self, data: object, value: int) -> None:
+        """
+        Check that an integer value fits in this data type's range.
+
+        Parameters
+        ----------
+        data : object
+            The original input, used in the error message.
+        value : int
+            The integer value to bounds-check.
+
+        Raises
+        ------
+        ValueError
+            If ``value`` is outside the range of the native dtype.
+        """
+        info = np.iinfo(self.to_native_dtype())
+        if value < info.min or value > info.max:
+            raise ValueError(f"{data!r} is out of bounds for {self.to_native_dtype()}.")
+
     def cast_scalar(self, data: object) -> Scalar:
         """
         Attempt to cast a given object to a NumPy integer scalar.
@@ -196,9 +216,12 @@ def cast_scalar(self, data: object) -> Scalar:
         ------
         TypeError
             If the data cannot be converted to a NumPy integer scalar.
+        ValueError
+            If the data is an integer outside the range of the dtype.
         """
 
         if self._check_scalar(data):
+            self._check_int_bounds(data, int(data))
             return self._cast_scalar_unchecked(data)
         msg = (
             f"Cannot convert object {data!r} with type {type(data)} to a scalar compatible with the "
@@ -237,16 +260,16 @@ def from_json_scalar(self, data: JSON, *, zarr_format: ZarrFormat) -> Scalar:
         ------
         TypeError
             If the input is not a valid integer type.
+        ValueError
+            If the input is an integer outside the range of the dtype.
         """
-        if check_json_int(data):
-            return self._cast_scalar_unchecked(data)
-        if check_json_intish_float(data):
-            return self._cast_scalar_unchecked(int(data))
-
-        if check_json_intish_str(data):
-            return self._cast_scalar_unchecked(int(data))
+        if check_json_int(data) or check_json_intish_float(data) or check_json_intish_str(data):
+            value = int(data)
+        else:
+            raise TypeError(f"Invalid type: {data}. Expected an integer.")
 
-        raise TypeError(f"Invalid type: {data}. Expected an integer.")
+        self._check_int_bounds(data, value)
+        return self._cast_scalar_unchecked(value)
 
     def to_json_scalar(self, data: object, *, zarr_format: ZarrFormat) -> int:
         """

tests/test_dtype/test_npy/test_int.py

@@ -1,6 +1,6 @@
 from __future__ import annotations
 
-from typing import Any
+from typing import TYPE_CHECKING, Any
 
 import numpy as np
 import pytest
@@ -10,6 +10,9 @@
 from zarr.core.dtype.npy.int import Int8, Int16, Int32, Int64, UInt8, UInt16, UInt32, UInt64
 from zarr.errors import DataTypeValidationError
 
+if TYPE_CHECKING:
+    from zarr.core.common import JSON
+
 
 class TestInt8(BaseTestZDType):
     test_cls = Int8
@@ -365,3 +368,56 @@ def test_match_dtype_not_a_numpy_dtype(not_a_dtype: Any) -> None:
     """
     with pytest.raises(ValueError, match="No Zarr data type found"):
         data_type_registry.match_dtype(not_a_dtype)
+
+
+@pytest.mark.parametrize(
+    ("dtype_instance", "value"),
+    [
+        (Int8(), 300),
+        (Int8(), -129),
+        (Int8(), "300"),
+        (Int8(), 300.0),
+        (UInt8(), -1),
+        (UInt64(), 2**64),
+        (Int64(), -(2**63) - 1),
+    ],
+)
+def test_out_of_bounds_integer_from_json_scalar(
+    dtype_instance: Int8 | UInt8 | UInt64 | Int64, value: JSON
+) -> None:
+    """An integer outside the dtype's range raises a validation error, not
+    numpy's OverflowError. Regression test for
+    https://github.com/zarr-developers/zarr-python/issues/4453 item 4."""
+    for zarr_format in (2, 3):
+        with pytest.raises(ValueError, match="out of bounds"):
+            dtype_instance.from_json_scalar(value, zarr_format=zarr_format)
+
+
+@pytest.mark.parametrize(
+    ("dtype_instance", "value"),
+    [
+        (Int8(), 127),
+        (Int8(), -128),
+        (UInt8(), 255),
+        (UInt64(), 2**64 - 1),
+        (Int64(), -(2**63)),
+    ],
+)
+def test_in_bounds_integer_from_json_scalar(
+    dtype_instance: Int8 | UInt8 | UInt64 | Int64, value: int
+) -> None:
+    """In-bounds integers round-trip for both zarr formats."""
+    for zarr_format in (2, 3):
+        assert dtype_instance.from_json_scalar(
+            value, zarr_format=zarr_format
+        ) == dtype_instance.to_native_dtype().type(value)
+
+
+def test_out_of_bounds_integer_cast_scalar() -> None:
+    """cast_scalar raises ValueError for out-of-bounds integers, so
+    create_array(dtype="i1", fill_value=300) fails the same way reading
+    stored metadata does."""
+    with pytest.raises(ValueError, match="out of bounds"):
+        Int8().cast_scalar(300)
+    with pytest.raises(ValueError, match="out of bounds"):
+        UInt64().cast_scalar(2**64)