Much nicer normalize_token for pd.RangeIndex

Author: dcherian

flox/dask_array_ops.py

@@ -4,14 +4,22 @@
 from itertools import product
 from numbers import Integral
 
+import pandas as pd
 from dask import config
+from dask.base import normalize_token
 from dask.blockwise import lol_tuples
 from toolz import partition_all
 
 from .lib import ArrayLayer
 from .types import Graph
 
 
+# workaround for https://github.com/dask/dask/issues/11862
+@normalize_token.register(pd.RangeIndex)
+def normalize_range_index(x):
+    return normalize_token(type(x)), x.start, x.stop, x.step, x.dtype, x.name
+
+
 # _tree_reduce and partial_reduce are copied from dask.array.reductions
 # They have been modified to work purely with graphs, and without creating new Array layers
 # in the graph. The `block_index` kwarg is new and avoids a concatenation by simply setting the right