BUG: non-64-bit numeric dtypes should raise in IntervalDtype constructor

Author: topper-123

doc/source/whatsnew/v2.0.0.rst

@@ -1237,6 +1237,8 @@ Interval
 - Bug in :meth:`IntervalIndex.is_overlapping` incorrect output if interval has duplicate left boundaries (:issue:`49581`)
 - Bug in :meth:`Series.infer_objects` failing to infer :class:`IntervalDtype` for an object series of :class:`Interval` objects (:issue:`50090`)
 - Bug in :meth:`Series.shift` with :class:`IntervalDtype` and invalid null ``fill_value`` failing to raise ``TypeError`` (:issue:`51258`)
+- Bug in :class:`IntervalDtype` where it accepted non-64-bit numeric subtypes, even though :class:`arrays.InterArray` only can hold numeric data if it is 64-bit.
+ Supplying non-64-bit numeric subtypes to :class:`IntervalDtype` now raises a ``TypeError`` (:issue:`45412`)
 -
 
 Indexing

pandas/core/arrays/interval.py

@@ -270,6 +270,7 @@ def __new__(
  copy=copy,
  dtype=dtype,
  )
+ 1 / 0
 
  if verify_integrity:
  cls._validate(left, right, dtype=dtype)

pandas/core/dtypes/common.py

@@ -861,6 +861,56 @@ def is_int64_dtype(arr_or_dtype) -> bool:
  return _is_dtype_type(arr_or_dtype, classes(np.int64))
 
 
+def is_64bit_real_numeric_dtype(arr_or_dtype) -> bool:
+ """
+ Check whether the provided array or dtype is of 64-bit dtype.
+
+ Parameters
+ ----------
+ arr_or_dtype : array-like or dtype
+ The array or dtype to check.
+
+ Returns
+ -------
+ boolean
+ Whether or not the array or dtype is of 64-bit dtype.
+
+ Examples
+ --------
+ >>> is_64bit_real_numeric_dtype(str)
+ False
+ >>> is_64bit_real_numeric_dtype(np.int32)
+ False
+ >>> is_64bit_real_numeric_dtype(np.int64)
+ True
+ >>> is_64bit_real_numeric_dtype('int8')
+ False
+ >>> is_64bit_real_numeric_dtype('Int8')
+ True
+ >>> is_64bit_real_numeric_dtype(pd.Int64Dtype)
+ True
+ >>> is_64bit_real_numeric_dtype(float)
+ True
+ >>> is_64bit_real_numeric_dtype(np.uint64) # unsigned
+ True
+ >>> is_64bit_real_numeric_dtype(np.array(['a', 'b']))
+ False
+ >>> is_64bit_real_numeric_dtype(np.array([1, 2], dtype=np.int64))
+ True
+ >>> is_64bit_real_numeric_dtype(pd.Index([1, 2.])) # float
+ True
+ >>> is_64bit_real_numeric_dtype(np.array([1, 2], dtype=np.uint32))
+ False
+ """
+ return _is_dtype_type(
+ arr_or_dtype, classes(np.int64, np.uint64, np.float64)
+ ) or _is_dtype(
+ arr_or_dtype,
+ lambda typ: isinstance(typ, ExtensionDtype)
+ and typ.type in (np.int64, np.uint64, np.float64),
+ )
+
+
 def is_datetime64_any_dtype(arr_or_dtype) -> bool:
  """
  Check whether the provided array or dtype is of the datetime64 dtype.

pandas/core/dtypes/dtypes.py

@@ -1082,6 +1082,8 @@ class IntervalDtype(PandasExtensionDtype):
 
  def __new__(cls, subtype=None, closed: str_type | None = None):
  from pandas.core.dtypes.common import (
+ is_64bit_real_numeric_dtype,
+ is_any_real_numeric_dtype,
  is_string_dtype,
  pandas_dtype,
  )
@@ -1132,6 +1134,12 @@ def __new__(cls, subtype=None, closed: str_type | None = None):
  "for IntervalDtype"
  )
  raise TypeError(msg)
+ elif is_any_real_numeric_dtype(subtype) and not is_64bit_real_numeric_dtype(
+ subtype
+ ):
+ raise TypeError(
+ f"numeric subtype must be 64-bit numeric dtype, was {subtype}"
+ )
 
  key = f"{subtype}{closed}"
  try:

pandas/tests/dtypes/test_dtypes.py

@@ -8,6 +8,7 @@
 
 from pandas.core.dtypes.base import _registry as registry
 from pandas.core.dtypes.common import (
+ is_64bit_real_numeric_dtype,
  is_bool_dtype,
  is_categorical_dtype,
  is_datetime64_any_dtype,
@@ -16,6 +17,7 @@
  is_datetime64tz_dtype,
  is_dtype_equal,
  is_interval_dtype,
+ is_numeric_dtype,
  is_period_dtype,
  is_string_dtype,
 )
@@ -597,6 +599,7 @@ def test_construction_generic(self, subtype):
  @pytest.mark.parametrize(
  "subtype",
  [
+ *[x for x in tm.ALL_REAL_DTYPES if not is_64bit_real_numeric_dtype(x)],
  CategoricalDtype(list("abc"), False),
  CategoricalDtype(list("wxyz"), True),
  object,
@@ -607,11 +610,14 @@ def test_construction_generic(self, subtype):
  ],
  )
  def test_construction_not_supported(self, subtype):
- # GH 19016
- msg = (
- "category, object, and string subtypes are not supported "
- "for IntervalDtype"
- )
+ # GH19016, GH45412
+ if is_numeric_dtype(subtype):
+ msg = "numeric subtype must be 64-bit numeric dtype, was"
+ else:
+ msg = (
+ "category, object, and string subtypes are not supported "
+ "for IntervalDtype"
+ )
  with pytest.raises(TypeError, match=msg):
  IntervalDtype(subtype)