BUG: silently ignoring dtype kwarg in Index.__new__ (pandas-dev#38879)

jbrockmendel · luckyvs1 · commit 120323647250 · 2021-01-19T23:18:36.000-08:00
diff --git a/doc/source/whatsnew/v1.3.0.rst b/doc/source/whatsnew/v1.3.0.rst
@@ -313,7 +313,7 @@ ExtensionArray
 
 Other
 ^^^^^
-
+- Bug in :class:`Index` constructor sometimes silently ignorning a a specified ``dtype`` (:issue:`38879`)
 -
 -
 
diff --git a/pandas/core/dtypes/common.py b/pandas/core/dtypes/common.py
@@ -1529,6 +1529,19 @@ def is_extension_array_dtype(arr_or_dtype) -> bool:
     return isinstance(dtype, ExtensionDtype) or registry.find(dtype) is not None
 
 
+def is_ea_or_datetimelike_dtype(dtype: Optional[DtypeObj]) -> bool:
+    """
+    Check for ExtensionDtype, datetime64 dtype, or timedelta64 dtype.
+
+    Notes
+    -----
+    Checks only for dtype objects, not dtype-castable strings or types.
+    """
+    return isinstance(dtype, ExtensionDtype) or (
+        isinstance(dtype, np.dtype) and dtype.kind in ["m", "M"]
+    )
+
+
 def is_complex_dtype(arr_or_dtype) -> bool:
     """
     Check whether the provided array or dtype is of a complex dtype.
diff --git a/pandas/core/indexes/base.py b/pandas/core/indexes/base.py
@@ -44,8 +44,8 @@
     ensure_platform_int,
     is_bool_dtype,
     is_categorical_dtype,
-    is_datetime64_any_dtype,
     is_dtype_equal,
+    is_ea_or_datetimelike_dtype,
     is_extension_array_dtype,
     is_float,
     is_float_dtype,
@@ -56,10 +56,8 @@
     is_iterator,
     is_list_like,
     is_object_dtype,
-    is_period_dtype,
     is_scalar,
     is_signed_integer_dtype,
-    is_timedelta64_dtype,
     is_unsigned_integer_dtype,
     needs_i8_conversion,
     pandas_dtype,
@@ -69,6 +67,7 @@
 from pandas.core.dtypes.dtypes import (
     CategoricalDtype,
     DatetimeTZDtype,
+    ExtensionDtype,
     IntervalDtype,
     PeriodDtype,
 )
@@ -87,6 +86,7 @@
 import pandas.core.algorithms as algos
 from pandas.core.arrays import Categorical, ExtensionArray
 from pandas.core.arrays.datetimes import tz_to_dtype, validate_tz_from_dtype
+from pandas.core.arrays.sparse import SparseDtype
 from pandas.core.base import IndexOpsMixin, PandasObject
 import pandas.core.common as com
 from pandas.core.construction import extract_array
@@ -286,44 +286,32 @@ def __new__(
 
         # range
         if isinstance(data, RangeIndex):
-            return RangeIndex(start=data, copy=copy, dtype=dtype, name=name)
+            result = RangeIndex(start=data, copy=copy, name=name)
+            if dtype is not None:
+                return result.astype(dtype, copy=False)
+            return result
         elif isinstance(data, range):
-            return RangeIndex.from_range(data, dtype=dtype, name=name)
-
-        # categorical
-        elif is_categorical_dtype(data_dtype) or is_categorical_dtype(dtype):
-            # Delay import for perf. https://github.com/pandas-dev/pandas/pull/31423
-            from pandas.core.indexes.category import CategoricalIndex
-
-            return _maybe_asobject(dtype, CategoricalIndex, data, copy, name, **kwargs)
-
-        # interval
-        elif is_interval_dtype(data_dtype) or is_interval_dtype(dtype):
-            # Delay import for perf. https://github.com/pandas-dev/pandas/pull/31423
-            from pandas.core.indexes.interval import IntervalIndex
-
-            return _maybe_asobject(dtype, IntervalIndex, data, copy, name, **kwargs)
-
-        elif is_datetime64_any_dtype(data_dtype) or is_datetime64_any_dtype(dtype):
-            # Delay import for perf. https://github.com/pandas-dev/pandas/pull/31423
-            from pandas import DatetimeIndex
-
-            return _maybe_asobject(dtype, DatetimeIndex, data, copy, name, **kwargs)
-
-        elif is_timedelta64_dtype(data_dtype) or is_timedelta64_dtype(dtype):
-            # Delay import for perf. https://github.com/pandas-dev/pandas/pull/31423
-            from pandas import TimedeltaIndex
-
-            return _maybe_asobject(dtype, TimedeltaIndex, data, copy, name, **kwargs)
-
-        elif is_period_dtype(data_dtype) or is_period_dtype(dtype):
-            # Delay import for perf. https://github.com/pandas-dev/pandas/pull/31423
-            from pandas import PeriodIndex
+            result = RangeIndex.from_range(data, name=name)
+            if dtype is not None:
+                return result.astype(dtype, copy=False)
+            return result
 
-            return _maybe_asobject(dtype, PeriodIndex, data, copy, name, **kwargs)
+        if is_ea_or_datetimelike_dtype(dtype):
+            # non-EA dtype indexes have special casting logic, so we punt here
+            klass = cls._dtype_to_subclass(dtype)
+            if klass is not Index:
+                return klass(data, dtype=dtype, copy=copy, name=name, **kwargs)
+
+        if is_ea_or_datetimelike_dtype(data_dtype):
+            klass = cls._dtype_to_subclass(data_dtype)
+            if klass is not Index:
+                result = klass(data, copy=copy, name=name, **kwargs)
+                if dtype is not None:
+                    return result.astype(dtype, copy=False)
+                return result
 
         # extension dtype
-        elif is_extension_array_dtype(data_dtype) or is_extension_array_dtype(dtype):
+        if is_extension_array_dtype(data_dtype) or is_extension_array_dtype(dtype):
             if not (dtype is None or is_object_dtype(dtype)):
                 # coerce to the provided dtype
                 ea_cls = dtype.construct_array_type()
@@ -407,26 +395,38 @@ def _ensure_array(cls, data, dtype, copy: bool):
     def _dtype_to_subclass(cls, dtype: DtypeObj):
         # Delay import for perf. https://github.com/pandas-dev/pandas/pull/31423
 
-        if isinstance(dtype, DatetimeTZDtype) or dtype == np.dtype("M8[ns]"):
+        if isinstance(dtype, ExtensionDtype):
+            if isinstance(dtype, DatetimeTZDtype):
+                from pandas import DatetimeIndex
+
+                return DatetimeIndex
+            elif isinstance(dtype, CategoricalDtype):
+                from pandas import CategoricalIndex
+
+                return CategoricalIndex
+            elif isinstance(dtype, IntervalDtype):
+                from pandas import IntervalIndex
+
+                return IntervalIndex
+            elif isinstance(dtype, PeriodDtype):
+                from pandas import PeriodIndex
+
+                return PeriodIndex
+
+            elif isinstance(dtype, SparseDtype):
+                return cls._dtype_to_subclass(dtype.subtype)
+
+            return Index
+
+        if dtype.kind == "M":
             from pandas import DatetimeIndex
 
             return DatetimeIndex
-        elif dtype == "m8[ns]":
+
+        elif dtype.kind == "m":
             from pandas import TimedeltaIndex
 
             return TimedeltaIndex
-        elif isinstance(dtype, CategoricalDtype):
-            from pandas import CategoricalIndex
-
-            return CategoricalIndex
-        elif isinstance(dtype, IntervalDtype):
-            from pandas import IntervalIndex
-
-            return IntervalIndex
-        elif isinstance(dtype, PeriodDtype):
-            from pandas import PeriodIndex
-
-            return PeriodIndex
 
         elif is_float_dtype(dtype):
             from pandas import Float64Index
@@ -445,6 +445,9 @@ def _dtype_to_subclass(cls, dtype: DtypeObj):
             # NB: assuming away MultiIndex
             return Index
 
+        elif issubclass(dtype.type, (str, bool, np.bool_)):
+            return Index
+
         raise NotImplementedError(dtype)
 
     """
@@ -6253,43 +6256,6 @@ def _try_convert_to_int_array(
     raise ValueError
 
 
-def _maybe_asobject(dtype, klass, data, copy: bool, name: Label, **kwargs):
-    """
-    If an object dtype was specified, create the non-object Index
-    and then convert it to object.
-
-    Parameters
-    ----------
-    dtype : np.dtype, ExtensionDtype, str
-    klass : Index subclass
-    data : list-like
-    copy : bool
-    name : hashable
-    **kwargs
-
-    Returns
-    -------
-    Index
-
-    Notes
-    -----
-    We assume that calling .astype(object) on this klass will make a copy.
-    """
-
-    # GH#23524 passing `dtype=object` to DatetimeIndex is invalid,
-    #  will raise in the where `data` is already tz-aware.  So
-    #  we leave it out of this step and cast to object-dtype after
-    #  the DatetimeIndex construction.
-
-    if is_dtype_equal(_o_dtype, dtype):
-        # Note we can pass copy=False because the .astype below
-        #  will always make a copy
-        index = klass(data, copy=False, name=name, **kwargs)
-        return index.astype(object)
-
-    return klass(data, dtype=dtype, copy=copy, name=name, **kwargs)
-
-
 def get_unanimous_names(*indexes: Index) -> Tuple[Label, ...]:
     """
     Return common name if all indices agree, otherwise None (level-by-level).
diff --git a/pandas/tests/indexes/ranges/test_constructors.py b/pandas/tests/indexes/ranges/test_constructors.py
@@ -114,11 +114,6 @@ def test_constructor_range(self):
         expected = RangeIndex(1, 5, 2)
         tm.assert_index_equal(result, expected, exact=True)
 
-        with pytest.raises(
-            ValueError,
-            match="Incorrect `dtype` passed: expected signed integer, received float64",
-        ):
-            Index(range(1, 5, 2), dtype="float64")
         msg = r"^from_range\(\) got an unexpected keyword argument"
         with pytest.raises(TypeError, match=msg):
             RangeIndex.from_range(range(10), copy=True)
diff --git a/pandas/tests/indexes/test_index_new.py b/pandas/tests/indexes/test_index_new.py
@@ -8,18 +8,22 @@
 
 from pandas import (
     NA,
+    Categorical,
     CategoricalIndex,
     DatetimeIndex,
     Index,
     Int64Index,
+    IntervalIndex,
     MultiIndex,
     NaT,
     PeriodIndex,
     Series,
     TimedeltaIndex,
     Timestamp,
     UInt64Index,
+    date_range,
     period_range,
+    timedelta_range,
 )
 import pandas._testing as tm
 
@@ -122,6 +126,80 @@ def test_constructor_mixed_nat_objs_infers_object(self, swap_objs):
         tm.assert_index_equal(Index(np.array(data, dtype=object)), expected)
 
 
+class TestDtypeEnforced:
+    # check we don't silently ignore the dtype keyword
+
+    @pytest.mark.parametrize("dtype", [object, "float64", "uint64", "category"])
+    def test_constructor_range_values_mismatched_dtype(self, dtype):
+        rng = Index(range(5))
+
+        result = Index(rng, dtype=dtype)
+        assert result.dtype == dtype
+
+        result = Index(range(5), dtype=dtype)
+        assert result.dtype == dtype
+
+    @pytest.mark.parametrize("dtype", [object, "float64", "uint64", "category"])
+    def test_constructor_categorical_values_mismatched_non_ea_dtype(self, dtype):
+        cat = Categorical([1, 2, 3])
+
+        result = Index(cat, dtype=dtype)
+        assert result.dtype == dtype
+
+    def test_constructor_categorical_values_mismatched_dtype(self):
+        dti = date_range("2016-01-01", periods=3)
+        cat = Categorical(dti)
+        result = Index(cat, dti.dtype)
+        tm.assert_index_equal(result, dti)
+
+        dti2 = dti.tz_localize("Asia/Tokyo")
+        cat2 = Categorical(dti2)
+        result = Index(cat2, dti2.dtype)
+        tm.assert_index_equal(result, dti2)
+
+        ii = IntervalIndex.from_breaks(range(5))
+        cat3 = Categorical(ii)
+        result = Index(cat3, dtype=ii.dtype)
+        tm.assert_index_equal(result, ii)
+
+    def test_constructor_ea_values_mismatched_categorical_dtype(self):
+        dti = date_range("2016-01-01", periods=3)
+        result = Index(dti, dtype="category")
+        expected = CategoricalIndex(dti)
+        tm.assert_index_equal(result, expected)
+
+        dti2 = date_range("2016-01-01", periods=3, tz="US/Pacific")
+        result = Index(dti2, dtype="category")
+        expected = CategoricalIndex(dti2)
+        tm.assert_index_equal(result, expected)
+
+    def test_constructor_period_values_mismatched_dtype(self):
+        pi = period_range("2016-01-01", periods=3, freq="D")
+        result = Index(pi, dtype="category")
+        expected = CategoricalIndex(pi)
+        tm.assert_index_equal(result, expected)
+
+    def test_constructor_timedelta64_values_mismatched_dtype(self):
+        # check we don't silently ignore the dtype keyword
+        tdi = timedelta_range("4 Days", periods=5)
+        result = Index(tdi, dtype="category")
+        expected = CategoricalIndex(tdi)
+        tm.assert_index_equal(result, expected)
+
+    def test_constructor_interval_values_mismatched_dtype(self):
+        dti = date_range("2016-01-01", periods=3)
+        ii = IntervalIndex.from_breaks(dti)
+        result = Index(ii, dtype="category")
+        expected = CategoricalIndex(ii)
+        tm.assert_index_equal(result, expected)
+
+    def test_constructor_datetime64_values_mismatched_period_dtype(self):
+        dti = date_range("2016-01-01", periods=3)
+        result = Index(dti, dtype="Period[D]")
+        expected = dti.to_period("D")
+        tm.assert_index_equal(result, expected)
+
+
 class TestIndexConstructorUnwrapping:
     # Test passing different arraylike values to pd.Index
 

Original file line number	Diff line number	Diff line change
`@@ -313,7 +313,7 @@ ExtensionArray`
`313`	`313`
`314`	`314`	`Other`
`315`	`315`	`^^^^^`
`316`		`-`
	`316`	+- Bug in :class:`Index` constructor sometimes silently ignorning a a specified ``dtype`` (:issue:`38879`)
`317`	`317`	`-`
`318`	`318`	`-`
`319`	`319`