Splitattrs equalise

lib/iris/common/_split_attribute_dicts.py

@@ -18,7 +18,6 @@
 So, we simply treat "global" and "local" attributes of the same name as entirely
 independent. Which happily is also the easiest to code, and to explain.
 """
-
 from collections.abc import Mapping, Sequence
 from functools import wraps
 
@@ -30,7 +29,16 @@ def _convert_splitattrs_to_pairedkeys_dict(dic):
     Transform a :class:`~iris.cube.CubeAttributesDict` "split" attributes dictionary
     into a 'normal' :class:`dict`, with paired keys of the form ('global', name) or
     ('local', name).
+
+    If the input is *not* a split-attrs dict, it is converted to one before
+    transforming it.  This will assign its keys to global/local depending on a standard
+    set of choices (see :class:`~iris.cube.CubeAttributesDict`).
     """
+    from iris.cube import CubeAttrsDict
+
+    # Convert input to CubeAttrsDict
+    if not hasattr(dic, "globals") or not hasattr(dic, "locals"):
+        dic = CubeAttrsDict(dic)
 
     def _global_then_local_items(dic):
         # Routine to produce global, then local 'items' in order, and with all keys
@@ -93,13 +101,6 @@ def adjust_for_split_attribute_dictionaries(operation):
 
     @wraps(operation)
     def _inner_function(*args, **kwargs):
-        from iris.cube import CubeAttrsDict
-
-        # First make all inputs into CubeAttrsDict, if not already.
-        args = [
-            arg if isinstance(arg, CubeAttrsDict) else CubeAttrsDict(arg)
-            for arg in args
-        ]
         # Convert all inputs into 'pairedkeys' type dicts
         args = [_convert_splitattrs_to_pairedkeys_dict(arg) for arg in args]
 

lib/iris/tests/unit/util/test_equalise_attributes.py

@@ -14,8 +14,13 @@
 
 import numpy as np
 
-from iris.cube import Cube
+from iris.coords import AuxCoord
+from iris.cube import Cube, CubeAttrsDict
 import iris.tests.stock
+from iris.tests.unit.common.metadata.test_CubeMetadata import (
+    _TEST_ATTRNAME,
+    make_attrsdict,
+)
 from iris.util import equalise_attributes
 
 
@@ -153,5 +158,111 @@ def test_complex_somecommon(self):
         )
 
 
+class TestSplitattributes:
+    """
+    Extra testing for cases where attributes differ specifically by type
+
+    That is, where there is a new possibility of 'mismatch' due to the newer "typing"
+    of attributes as global or local.
+
+    Specifically, it is now possible that although
+    "cube1.attributes.keys == cube2.attributes.keys()",
+    AND "cube1.attributes[k] == cube2.attributes[k]" for all keys,
+    YET STILL (possibly) "cube1.attributes != cube2.attributes"
+    """
+
+    @staticmethod
+    def _sample_splitattrs_cube(attr_global_local):
+        attrs = CubeAttrsDict(
+            globals=make_attrsdict(attr_global_local[0]),
+            locals=make_attrsdict(attr_global_local[1]),
+        )
+        return Cube([0], attributes=attrs)
+
+    @staticmethod
+    def check_equalised_result(cube1, cube2):
+        equalise_attributes([cube1, cube2])
+        # Note: "X" represents a missing attribute, as in test_CubeMetadata
+        return [
+            (
+                cube1.attributes.globals.get(_TEST_ATTRNAME, "X")
+                + cube1.attributes.locals.get(_TEST_ATTRNAME, "X")
+            ),
+            (
+                cube2.attributes.globals.get(_TEST_ATTRNAME, "X")
+                + cube2.attributes.locals.get(_TEST_ATTRNAME, "X")
+            ),
+        ]
+
+    def test__global_and_local__bothsame(self):
+        # A trivial case showing that the original globals+locals are both preserved.
+        cube1 = self._sample_splitattrs_cube("AB")
+        cube2 = self._sample_splitattrs_cube("AB")
+        result = self.check_equalised_result(cube1, cube2)
+        assert result == ["AB", "AB"]
+
+    def test__globals_different(self):
+        cube1 = self._sample_splitattrs_cube("AX")
+        cube2 = self._sample_splitattrs_cube("BX")
+        result = self.check_equalised_result(cube1, cube2)
+        assert result == ["XX", "XX"]
+
+    def test__locals_different(self):
+        cube1 = self._sample_splitattrs_cube("XA")
+        cube2 = self._sample_splitattrs_cube("XB")
+        result = self.check_equalised_result(cube1, cube2)
+        assert result == ["XX", "XX"]
+
+    def test__oneglobal_onelocal__different(self):
+        cube1 = self._sample_splitattrs_cube("AX")
+        cube2 = self._sample_splitattrs_cube("XB")
+        result = self.check_equalised_result(cube1, cube2)
+        assert result == ["XX", "XX"]
+
+    # This case fails without the split-attributes fix.
+    def test__oneglobal_onelocal__same(self):
+        cube1 = self._sample_splitattrs_cube("AX")
+        cube2 = self._sample_splitattrs_cube("XA")
+        result = self.check_equalised_result(cube1, cube2)
+        assert result == ["XX", "XX"]
+
+    def test__sameglobals_onelocal__different(self):
+        cube1 = self._sample_splitattrs_cube("AB")
+        cube2 = self._sample_splitattrs_cube("AX")
+        result = self.check_equalised_result(cube1, cube2)
+        assert result == ["XX", "XX"]
+
+    # This case fails without the split-attributes fix.
+    def test__sameglobals_onelocal__same(self):
+        cube1 = self._sample_splitattrs_cube("AA")
+        cube2 = self._sample_splitattrs_cube("AX")
+        result = self.check_equalised_result(cube1, cube2)
+        assert result == ["XX", "XX"]
+
+    # This case fails without the split-attributes fix.
+    def test__differentglobals_samelocals(self):
+        cube1 = self._sample_splitattrs_cube("AC")
+        cube2 = self._sample_splitattrs_cube("BC")
+        result = self.check_equalised_result(cube1, cube2)
+        assert result == ["XX", "XX"]
+
+
+class TestNonCube:
+    # Just to assert that we can do operations on non-cube components (like Coords),
+    # in fact effectively, anything with a ".attributes".
+    # Even though the docstring does not admit this, we test it because we put in
+    # special code to preserve it when adding the split-attribute handling.
+    def test(self):
+        attrs = [1, 1, 2]
+        coords = [
+            AuxCoord([0], attributes={"a": attr, "b": "all_the_same"})
+            for attr in attrs
+        ]
+        equalise_attributes(coords)
+        assert all(
+            coord.attributes == {"b": "all_the_same"} for coord in coords
+        )
+
+
 if __name__ == "__main__":
     tests.main()

lib/iris/util.py

@@ -2059,31 +2059,53 @@ def equalise_attributes(cubes):
     See more at :doc:`/userguide/real_and_lazy_data`.
 
     """
+    # deferred imports to avoid circularity problem
+    from iris.common._split_attribute_dicts import (
+        _convert_splitattrs_to_pairedkeys_dict,
+    )
+    from iris.cube import CubeAttrsDict
+
+    cube_attrs = [cube.attributes for cube in cubes]
     removed = []
+
+    # *IF* they are cube attributes, then convert to "paired-keys" form.
+    is_split_dicts = any(
+        isinstance(attrs, CubeAttrsDict) for attrs in cube_attrs
+    )
+    if is_split_dicts:
+        # Note: we only test this in case someone was passing some *other* objects for
+        # attributes processing (though the docstring does not admit this is possible).
+        cube_attrs = [
+            _convert_splitattrs_to_pairedkeys_dict(dic) for dic in cube_attrs
+        ]
+
     # Work out which attributes are identical across all the cubes.
-    common_keys = list(cubes[0].attributes.keys())
+    common_keys = list(cube_attrs[0].keys())
     keys_to_remove = set(common_keys)
-    for cube in cubes[1:]:
-        cube_keys = list(cube.attributes.keys())
+    for attrs in cube_attrs[1:]:
+        cube_keys = list(attrs.keys())
         keys_to_remove.update(cube_keys)
         common_keys = [
             key
             for key in common_keys
-            if (
-                key in cube_keys
-                and np.all(cube.attributes[key] == cubes[0].attributes[key])
-            )
+            if (key in cube_keys and np.all(attrs[key] == cube_attrs[0][key]))
         ]
     keys_to_remove.difference_update(common_keys)
 
-    # Remove all the other attributes.
+    # rework the keys list if we were handling split-attributes.
+    if is_split_dicts:
+        # Collect all the ('local'/'global, name) keys + retain just the names.
+        keys_to_remove = set(key_pair[1] for key_pair in keys_to_remove)
+
+    # Remove all the non-matching attributes.
     for cube in cubes:
         deleted_attributes = {
             key: cube.attributes.pop(key)
             for key in keys_to_remove
             if key in cube.attributes
         }
         removed.append(deleted_attributes)
+
     return removed