Fix IntervalDtype Bugs and Inconsistencies

Author: jschendel

doc/source/whatsnew/v0.23.0.txt

@@ -270,6 +270,7 @@ Other API Changes
 - Subtraction of :class:`Series` with timezone-aware ``dtype='datetime64[ns]'`` with mis-matched timezones will raise ``TypeError`` instead of ``ValueError`` (issue:`18817`)
 - :class:`IntervalIndex` and ``IntervalDtype`` no longer support categorical, object, and string subtypes (:issue:`19016`)
 - The default ``Timedelta`` constructor now accepts an ``ISO 8601 Duration`` string as an argument (:issue:`19040`)
+- ``IntervalDtype`` now returns ``True`` when compared against ``'interval'`` regardless of subtype, and ``IntervalDtype.name`` now returns ``'interval'`` regardless of subtype (:issue:`18980`)
 
 .. _whatsnew_0230.deprecations:
 
@@ -377,7 +378,7 @@ Conversion
 - Fixed bug where comparing :class:`DatetimeIndex` failed to raise ``TypeError`` when attempting to compare timezone-aware and timezone-naive datetimelike objects (:issue:`18162`)
 - Bug in :class:`DatetimeIndex` where the repr was not showing high-precision time values at the end of a day (e.g., 23:59:59.999999999) (:issue:`19030`)
 - Bug where dividing a scalar timedelta-like object with :class:`TimedeltaIndex` performed the reciprocal operation (:issue:`19125`)
--
+- Bug in ``IntervalDtype`` when constructing two instances with subtype ``CategoricalDtype`` where the second instance used cached attributes from the first (:issue:`18980`)
 
 Indexing
 ^^^^^^^^

pandas/core/dtypes/dtypes.py

@@ -626,6 +626,7 @@ class IntervalDtype(ExtensionDtype):
 
     THIS IS NOT A REAL NUMPY DTYPE
     """
+    name = 'interval'
     type = IntervalDtypeType
     kind = None
     str = '|O08'
@@ -653,8 +654,8 @@ def __new__(cls, subtype=None):
             u.subtype = None
             return u
         elif (isinstance(subtype, compat.string_types) and
-              subtype == 'interval'):
-            subtype = ''
+              subtype in ('interval', 'interval[]')):
+            subtype = None
         else:
             if isinstance(subtype, compat.string_types):
                 m = cls._match.search(subtype)
@@ -678,11 +679,15 @@ def __new__(cls, subtype=None):
             raise TypeError(msg)
 
         try:
-            return cls._cache[str(subtype)]
+            # GH 18980: need to combine since str and hash individually may not
+            # be unique, e.g. str(CategoricalDtype) always returns 'category',
+            # and hash(np.dtype('<m8')) == hash(np.dtype('<m8[ns]'))
+            key = ''.join([str(subtype), str(hash(subtype))])
+            return cls._cache[key]
         except KeyError:
             u = object.__new__(cls)
             u.subtype = subtype
-            cls._cache[str(subtype)] = u
+            cls._cache[key] = u
             return u
 
     @classmethod
@@ -692,31 +697,29 @@ def construct_from_string(cls, string):
         if its not possible
         """
         if isinstance(string, compat.string_types):
-            try:
-                return cls(string)
-            except ValueError:
-                pass
-        raise TypeError("could not construct IntervalDtype")
+            return cls(string)
+        msg = "a string needs to be passed, got type {typ}"
+        raise TypeError(msg.format(typ=type(string)))
 
     def __unicode__(self):
         if self.subtype is None:
             return "interval"
         return "interval[{subtype}]".format(subtype=self.subtype)
 
-    @property
-    def name(self):
-        return str(self)
-
     def __hash__(self):
         # make myself hashable
         return hash(str(self))
 
     def __eq__(self, other):
         if isinstance(other, compat.string_types):
-            return other == self.name or other == self.name.title()
-
-        return (isinstance(other, IntervalDtype) and
-                self.subtype == other.subtype)
+            return other.title() in (self.name.title(), str(self).title())
+        elif not isinstance(other, IntervalDtype):
+            return False
+        elif self.subtype is None or other.subtype is None:
+            # None should match any subtype
+            return True
+        else:
+            return self.subtype == other.subtype
 
     @classmethod
     def is_dtype(cls, dtype):

pandas/tests/dtypes/test_dtypes.py

@@ -433,7 +433,7 @@ def test_hash_vs_equality(self):
         assert dtype2 == dtype
         assert dtype3 == dtype
         assert dtype is dtype2
-        assert dtype2 is dtype
+        assert dtype2 is dtype3
         assert dtype3 is dtype
         assert hash(dtype) == hash(dtype2)
         assert hash(dtype) == hash(dtype3)
@@ -451,14 +451,19 @@ def test_hash_vs_equality(self):
         assert hash(dtype2) == hash(dtype2)
         assert hash(dtype2) == hash(dtype3)
 
-    def test_construction(self):
-        with pytest.raises(ValueError):
-            IntervalDtype('xx')
+    @pytest.mark.parametrize('subtype', [
+        'interval[int64]', 'Interval[int64]', 'int64', np.dtype('int64')])
+    def test_construction(self, subtype):
+        i = IntervalDtype(subtype)
+        assert i.subtype == np.dtype('int64')
+        assert is_interval_dtype(i)
 
-        for s in ['interval[int64]', 'Interval[int64]', 'int64']:
-            i = IntervalDtype(s)
-            assert i.subtype == np.dtype('int64')
-            assert is_interval_dtype(i)
+    @pytest.mark.parametrize('subtype', [None, 'interval', 'interval[]'])
+    def test_construction_generic(self, subtype):
+        # generic
+        i = IntervalDtype(subtype)
+        assert i.subtype is None
+        assert is_interval_dtype(i)
 
     @pytest.mark.parametrize('subtype', [
         CategoricalDtype(list('abc'), False),
@@ -471,17 +476,27 @@ def test_construction_not_supported(self, subtype):
         with tm.assert_raises_regex(TypeError, msg):
             IntervalDtype(subtype)
 
-    def test_construction_generic(self):
-        # generic
-        i = IntervalDtype('interval')
-        assert i.subtype == ''
-        assert is_interval_dtype(i)
-        assert str(i) == 'interval[]'
+    def test_construction_errors(self):
+        msg = 'could not construct IntervalDtype'
+        with tm.assert_raises_regex(ValueError, msg):
+            IntervalDtype('xx')
 
-        i = IntervalDtype()
-        assert i.subtype is None
-        assert is_interval_dtype(i)
-        assert str(i) == 'interval'
+    def test_construction_from_string(self):
+        result = IntervalDtype('interval[int64]')
+        assert is_dtype_equal(self.dtype, result)
+        result = IntervalDtype.construct_from_string('interval[int64]')
+        assert is_dtype_equal(self.dtype, result)
+
+    @pytest.mark.parametrize('string', [
+        'foo', 'interval[foo]', 'foo[int64]', 0, 3.14, ('a', 'b'), None])
+    def test_construction_from_string_errors(self, string):
+        if isinstance(string, string_types):
+            error, msg = ValueError, 'could not construct IntervalDtype'
+        else:
+            error, msg = TypeError, 'a string needs to be passed, got type'
+
+        with tm.assert_raises_regex(error, msg):
+            IntervalDtype.construct_from_string(string)
 
     def test_subclass(self):
         a = IntervalDtype('interval[int64]')
@@ -506,36 +521,45 @@ def test_is_dtype(self):
         assert not IntervalDtype.is_dtype(np.int64)
         assert not IntervalDtype.is_dtype(np.float64)
 
-    def test_identity(self):
-        assert (IntervalDtype('interval[int64]') ==
-                IntervalDtype('interval[int64]'))
-
     def test_coerce_to_dtype(self):
         assert (_coerce_to_dtype('interval[int64]') ==
                 IntervalDtype('interval[int64]'))
 
-    def test_construction_from_string(self):
-        result = IntervalDtype('interval[int64]')
-        assert is_dtype_equal(self.dtype, result)
-        result = IntervalDtype.construct_from_string('interval[int64]')
-        assert is_dtype_equal(self.dtype, result)
-        with pytest.raises(TypeError):
-            IntervalDtype.construct_from_string('foo')
-        with pytest.raises(TypeError):
-            IntervalDtype.construct_from_string('interval[foo]')
-        with pytest.raises(TypeError):
-            IntervalDtype.construct_from_string('foo[int64]')
-
     def test_equality(self):
         assert is_dtype_equal(self.dtype, 'interval[int64]')
         assert is_dtype_equal(self.dtype, IntervalDtype('int64'))
-        assert is_dtype_equal(self.dtype, IntervalDtype('int64'))
         assert is_dtype_equal(IntervalDtype('int64'), IntervalDtype('int64'))
 
         assert not is_dtype_equal(self.dtype, 'int64')
         assert not is_dtype_equal(IntervalDtype('int64'),
                                   IntervalDtype('float64'))
 
+    @pytest.mark.parametrize('subtype', [
+        None, 'interval', 'interval[]', 'int64', 'uint64', 'float64', object,
+        CategoricalDtype(), 'datetime64', 'timedelta64', PeriodDtype('Q')])
+    def test_equality_generic(self, subtype):
+        # GH 18980
+        dtype = IntervalDtype(subtype)
+        assert is_dtype_equal(dtype, 'interval')
+        assert is_dtype_equal(dtype, IntervalDtype())
+
+    @pytest.mark.parametrize('subtype', [
+        'int64', 'uint64', 'float64', 'complex128', np.dtype('O'),
+        CategoricalDtype(), 'datetime64', 'timedelta64', PeriodDtype('Q')])
+    def test_name_repr(self, subtype):
+        # GH 18980
+        dtype = IntervalDtype(subtype)
+        expected = 'interval[{subtype}]'.format(subtype=subtype)
+        assert str(dtype) == expected
+        assert dtype.name == 'interval'
+
+    @pytest.mark.parametrize('subtype', [None, 'interval', 'interval[]'])
+    def test_name_repr_generic(self, subtype):
+        # GH 18980
+        dtype = IntervalDtype(subtype)
+        assert str(dtype) == 'interval'
+        assert dtype.name == 'interval'
+
     def test_basic(self):
         assert is_interval_dtype(self.dtype)
 
@@ -576,6 +600,15 @@ def test_caching(self):
         tm.round_trip_pickle(dtype)
         assert len(IntervalDtype._cache) == 0
 
+    def test_caching_categoricaldtype(self):
+        # GH 18980
+        cdt1 = CategoricalDtype(list('abc'), True)
+        cdt2 = CategoricalDtype(list('wxyz'), False)
+        idt1 = IntervalDtype(cdt1)
+        idt2 = IntervalDtype(cdt2)
+        assert idt1.subtype is cdt1
+        assert idt2.subtype is cdt2
+
 
 class TestCategoricalDtypeParametrized(object):