Fix merging when index name in meta missmatches actual name (#1119)

Co-authored-by: Hendrik Makait <hendrik@makait.com>

dask_expr/_shuffle.py

@@ -731,7 +731,9 @@ def _get_index(idx, obj):
  if idx.ndim == 1:
  idx = idx.to_frame()
  elif index_shuffle:
- idx = obj.index.to_frame()
+ # set default index name, otherwise we will end up with 0
+ name = {"name": "_index"} if idx == ["_index"] else {}
+ idx = obj.index.to_frame(**name)
  else:
  idx = _select_columns_or_index(obj, idx)
  return idx

dask_expr/tests/test_merge.py

@@ -2,8 +2,9 @@
 
 import numpy as np
 import pytest
+from dask import delayed
 
-from dask_expr import Merge, from_pandas, merge, repartition
+from dask_expr import Merge, from_delayed, from_pandas, merge, repartition
 from dask_expr._expr import Filter, Projection
 from dask_expr._merge import BroadcastJoin
 from dask_expr._shuffle import Shuffle
@@ -1031,6 +1032,22 @@ def test_merge_after_rename(index):
  assert_eq(result, expected, check_index=False)
 
 
+def test_mismatching_meta():
+ df1 = from_pandas(pd.DataFrame({"value": [1, 2, 3]}))
+
+ def creator(i):
+ return pd.DataFrame(
+ {"value2": [1, 2, 3]}, index=pd.Index([1, 2, 3], name="index")
+ )
+
+ df2 = from_delayed(
+ [delayed(creator)(i) for i in range(3)],
+ meta=pd.DataFrame({"value2": [1, 2, 3]}),
+ )
+ result = df1.join(df2)
+ assert_eq(result, df1.compute().join(df2.compute()), check_index=False)
+
+
 def test_merge_tuple_left_on():
  df = pd.DataFrame(
  {