Make func optional in reduction

cubed/core/groupby.py

@@ -71,7 +71,7 @@ def wrapper(a, by, **kwargs):
     # then reduce across blocks
     return reduction_new(
         out,
-        func=_identity_func,
+        func=None,
         combine_func=combine_func,
         aggregate_func=aggregate_func,
         axis=(dummy_axis, axis),  # dummy and group axis
@@ -82,8 +82,3 @@ def wrapper(a, by, **kwargs):
         combine_sizes={axis: num_groups},  # group axis doesn't have size 1
         extra_func_kwargs=dict(dtype=intermediate_dtype, dummy_axis=dummy_axis),
     )
-
-
-def _identity_func(a, **kwargs):
-    # pass through
-    return a

cubed/core/ops.py

@@ -1022,6 +1022,10 @@ def reduction_new(
 ) -> "Array":
     """Apply a function to reduce an array along one or more axes."""
     if combine_func is None:
+        if func is None:
+            raise ValueError(
+                "At least one of `func` and `combine_func` must be specified in reduction"
+            )
         combine_func = func
     if axis is None:
         axis = tuple(range(x.ndim))
@@ -1033,12 +1037,16 @@ def reduction_new(
 
     split_every = _normalize_split_every(split_every, axis)
 
+    if func is None:
+        initial_func = None
+    else:
+        initial_func = partial(
+            func, axis=axis, keepdims=True, **(extra_func_kwargs or {})
+        )
     result = partial_reduce(
         x,
         partial(combine_func, **(extra_func_kwargs or {})),
-        initial_func=partial(
-            func, axis=axis, keepdims=True, **(extra_func_kwargs or {})
-        ),
+        initial_func=initial_func,
         split_every=split_every,
         dtype=intermediate_dtype,
         combine_sizes=combine_sizes,