refactor indexing.py: introduce .oindex for Explicitly Indexed Arrays

xarray/core/indexing.py

@@ -325,6 +325,18 @@ def as_integer_slice(value):
     return slice(start, stop, step)
 
 
+class IndexCallable:
+    """Provide getitem syntax for a callable object."""
+
+    __slots__ = ("func",)
+
+    def __init__(self, func):
+        self.func = func
+
+    def __getitem__(self, key):
+        return self.func(key)
+
+
 class BasicIndexer(ExplicitIndexer):
     """Tuple for basic indexing.
 
@@ -470,6 +482,13 @@ def __array__(self, dtype: np.typing.DTypeLike = None) -> np.ndarray:
         # Note this is the base class for all lazy indexing classes
         return np.asarray(self.get_duck_array(), dtype=dtype)
 
+    def _oindex(self, key):
+        raise NotImplementedError("This method should be overridden")
+
+    @property
+    def oindex(self):
+        return IndexCallable(self._oindex)
+
 
 class ImplicitToExplicitIndexingAdapter(NDArrayMixin):
     """Wrap an array, converting tuples into the indicated explicit indexer."""
@@ -560,6 +579,9 @@ def get_duck_array(self):
     def transpose(self, order):
         return LazilyVectorizedIndexedArray(self.array, self.key).transpose(order)
 
+    def _oindex(self, indexer):
+        return type(self)(self.array, self._updated_key(indexer))
+
     def __getitem__(self, indexer):
         if isinstance(indexer, VectorizedIndexer):
             array = LazilyVectorizedIndexedArray(self.array, self.key)
@@ -663,6 +685,9 @@ def _ensure_copied(self):
     def get_duck_array(self):
         return self.array.get_duck_array()
 
+    def _oindex(self, key):
+        return type(self)(_wrap_numpy_scalars(self.array[key]))
+
     def __getitem__(self, key):
         return type(self)(_wrap_numpy_scalars(self.array[key]))
 
@@ -696,6 +721,9 @@ def get_duck_array(self):
         self._ensure_cached()
         return self.array.get_duck_array()
 
+    def _oindex(self, key):
+        return type(self)(_wrap_numpy_scalars(self.array[key]))
+
     def __getitem__(self, key):
         return type(self)(_wrap_numpy_scalars(self.array[key]))
 
@@ -1332,6 +1360,10 @@ def _indexing_array_and_key(self, key):
     def transpose(self, order):
         return self.array.transpose(order)
 
+    def _oindex(self, key):
+        array, key = self._indexing_array_and_key(key)
+        return array[key]
+
     def __getitem__(self, key):
         array, key = self._indexing_array_and_key(key)
         return array[key]
@@ -1376,16 +1408,19 @@ def __init__(self, array):
             )
         self.array = array
 
+    def _oindex(self, key):
+        # manual orthogonal indexing (implemented like DaskIndexingAdapter)
+        key = key.tuple
+        value = self.array
+        for axis, subkey in reversed(list(enumerate(key))):
+            value = value[(slice(None),) * axis + (subkey, Ellipsis)]
+        return value
+
     def __getitem__(self, key):
         if isinstance(key, BasicIndexer):
             return self.array[key.tuple]
         elif isinstance(key, OuterIndexer):
-            # manual orthogonal indexing (implemented like DaskIndexingAdapter)
-            key = key.tuple
-            value = self.array
-            for axis, subkey in reversed(list(enumerate(key))):
-                value = value[(slice(None),) * axis + (subkey, Ellipsis)]
-            return value
+            return self.oindex[key]
         else:
             if isinstance(key, VectorizedIndexer):
                 raise TypeError("Vectorized indexing is not supported")
@@ -1395,11 +1430,10 @@ def __getitem__(self, key):
     def __setitem__(self, key, value):
         if isinstance(key, (BasicIndexer, OuterIndexer)):
             self.array[key.tuple] = value
+        elif isinstance(key, VectorizedIndexer):
+            raise TypeError("Vectorized indexing is not supported")
         else:
-            if isinstance(key, VectorizedIndexer):
-                raise TypeError("Vectorized indexing is not supported")
-            else:
-                raise TypeError(f"Unrecognized indexer: {key}")
+            raise TypeError(f"Unrecognized indexer: {key}")
 
     def transpose(self, order):
         xp = self.array.__array_namespace__()
@@ -1417,24 +1451,25 @@ def __init__(self, array):
         """
         self.array = array
 
-    def __getitem__(self, key):
+    def _oindex(self, key):
+        key = key.tuple
+        try:
+            return self.array[key]
+        except NotImplementedError:
+            # manual orthogonal indexing
+            value = self.array
+            for axis, subkey in reversed(list(enumerate(key))):
+                value = value[(slice(None),) * axis + (subkey,)]
+            return value
 
+    def __getitem__(self, key):
         if isinstance(key, BasicIndexer):
             return self.array[key.tuple]
         elif isinstance(key, VectorizedIndexer):
             return self.array.vindex[key.tuple]
         else:
             assert isinstance(key, OuterIndexer)
-            key = key.tuple
-            try:
-                return self.array[key]
-            except NotImplementedError:
-                # manual orthogonal indexing.
-                # TODO: port this upstream into dask in a saner way.
-                value = self.array
-                for axis, subkey in reversed(list(enumerate(key))):
-                    value = value[(slice(None),) * axis + (subkey,)]
-                return value
+            return self.oindex[key]
 
     def __setitem__(self, key, value):
         if isinstance(key, BasicIndexer):
@@ -1510,6 +1545,9 @@ def _convert_scalar(self, item):
         # a NumPy array.
         return to_0d_array(item)
 
+    def _oindex(self, key):
+        return self.__getitem__(key)
+
     def __getitem__(
         self, indexer
     ) -> (

xarray/core/variable.py

@@ -41,11 +41,7 @@
     maybe_coerce_to_str,
 )
 from xarray.namedarray.core import NamedArray, _raise_if_any_duplicate_dimensions
-from xarray.namedarray.pycompat import (
-    integer_types,
-    is_0d_dask_array,
-    to_duck_array,
-)
+from xarray.namedarray.pycompat import integer_types, is_0d_dask_array, to_duck_array
 
 NON_NUMPY_SUPPORTED_ARRAY_TYPES = (
     indexing.ExplicitlyIndexed,
@@ -761,7 +757,14 @@ def __getitem__(self, key) -> Self:
         array `x.values` directly.
         """
         dims, indexer, new_order = self._broadcast_indexes(key)
-        data = as_indexable(self._data)[indexer]
+        indexable = as_indexable(self._data)
+
+        if isinstance(indexer, BasicIndexer):
+            data = indexable[indexer]
+        elif isinstance(indexer, OuterIndexer):
+            data = indexable.oindex[indexer]
+        else:
+            data = indexable[indexer]
         if new_order:
             data = np.moveaxis(data, range(len(new_order)), new_order)
         return self._finalize_indexing_result(dims, data)
@@ -794,7 +797,12 @@ def _getitem_with_mask(self, key, fill_value=dtypes.NA):
             else:
                 actual_indexer = indexer
 
-            data = as_indexable(self._data)[actual_indexer]
+            indexable = as_indexable(self._data)
+
+            if isinstance(indexer, OuterIndexer):
+                data = indexable.oindex[indexer]
+            else:
+                data = indexable[actual_indexer]
             mask = indexing.create_mask(indexer, self.shape, data)
             # we need to invert the mask in order to pass data first. This helps
             # pint to choose the correct unit