EA: require size instead of __len__

doc/source/whatsnew/v1.0.0.rst

@@ -123,6 +123,7 @@ source, you should no longer need to install Cython into your build environment
 Backwards incompatible API changes
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
+- The :class:`ExtensionArray` interface now requires the author to implement ``size`` instead of ``__len__``.  The ``size`` method must _not_ depend on either ``__len__`` or ``shape`` (:issue:`28389`)
 - :class:`pandas.core.groupby.GroupBy.transform` now raises on invalid operation names (:issue:`27489`).
 - :class:`pandas.core.arrays.IntervalArray` adopts a new ``__repr__`` in accordance with other array classes (:issue:`25022`)
 

pandas/core/arrays/base.py

@@ -83,7 +83,7 @@ class ExtensionArray:
     * _from_sequence
     * _from_factorized
     * __getitem__
-    * __len__
+    * size
     * dtype
     * nbytes
     * isna
@@ -319,7 +319,7 @@ def __len__(self) -> int:
         -------
         length : int
         """
-        raise AbstractMethodError(self)
+        return self.shape[0]
 
     def __iter__(self):
         """
@@ -342,19 +342,28 @@ def dtype(self) -> ExtensionDtype:
         """
         raise AbstractMethodError(self)
 
+    @property
+    def size(self) -> int:
+        """
+        An instance of 'ExtensionDtype'.
+
+        Must *not* depend on self.shape or self.__len__
+        """
+        raise AbstractMethodError(self)
+
     @property
     def shape(self) -> Tuple[int, ...]:
         """
         Return a tuple of the array dimensions.
         """
-        return (len(self),)
+        return (self.size,)
 
     @property
     def ndim(self) -> int:
         """
         Extension Arrays are only allowed to be 1-dimensional.
         """
-        return 1
+        return len(self.shape)
 
     @property
     def nbytes(self) -> int:

pandas/core/arrays/datetimelike.py

@@ -394,7 +394,7 @@ def __array__(self, dtype=None):
     @property
     def size(self) -> int:
         """The number of elements in this array."""
-        return np.prod(self.shape)
+        return self._data.size
 
     def __len__(self):
         return len(self._data)

pandas/core/arrays/integer.py

@@ -469,8 +469,9 @@ def __setitem__(self, key, value):
         self._data[key] = value
         self._mask[key] = mask
 
-    def __len__(self):
-        return len(self._data)
+    @property
+    def size(self) -> int:
+        return self._data.size
 
     @property
     def nbytes(self):

pandas/core/arrays/numpy_.py

@@ -240,8 +240,9 @@ def __setitem__(self, key, value):
 
         self._ndarray[key] = value
 
-    def __len__(self) -> int:
-        return len(self._ndarray)
+    @property
+    def size(self) -> int:
+        return self._ndarray.size
 
     @property
     def nbytes(self) -> int:

pandas/core/arrays/sparse/array.py

@@ -526,7 +526,8 @@ def _valid_sp_values(self):
         mask = notna(sp_vals)
         return sp_vals[mask]
 
-    def __len__(self) -> int:
+    @property
+    def size(self) -> int:
         return self.sp_index.length
 
     @property

pandas/tests/extension/arrow/arrays.py

@@ -88,7 +88,8 @@ def __getitem__(self, item):
             vals = self._data.to_pandas()[item]
             return type(self).from_scalars(vals)
 
-    def __len__(self):
+    @property
+    def size(self) -> int:
         return len(self._data)
 
     def astype(self, dtype, copy=True):

pandas/tests/extension/decimal/array.py

@@ -137,7 +137,8 @@ def __setitem__(self, key, value):
             value = decimal.Decimal(value)
         self._data[key] = value
 
-    def __len__(self) -> int:
+    @property
+    def size(self) -> int:
         return len(self._data)
 
     @property

pandas/tests/extension/json/array.py

@@ -106,7 +106,8 @@ def __setitem__(self, key, value):
                     assert isinstance(v, self.dtype.type)
                     self.data[k] = v
 
-    def __len__(self) -> int:
+    @property
+    def size(self) -> int:
         return len(self.data)
 
     @property