FIX-modin-project#6935: Fix Merge failed when right operand is an emp…

…ty dataframe (modin-project#6941)

Co-authored-by: Iaroslav Igoshev <Poolliver868@mail.ru>
Signed-off-by: arunjose696 <arunjose696@gmail.com>

modin/core/dataframe/pandas/dataframe/dataframe.py

@@ -3203,6 +3203,25 @@ def _prepare_frame_to_broadcast(self, axis, indices, broadcast_all):
             passed_len += len(internal)
         return result_dict
 
+    def _extract_partitions(self):
+        """
+        Extract partitions if partitions are present.
+
+        If partitions are empty return a dummy partition with empty data but
+        index and columns of current dataframe.
+
+        Returns
+        -------
+        np.ndarray
+            NumPy array with extracted partitions.
+        """
+        if self._partitions.size > 0:
+            return self._partitions
+        else:
+            return self._partition_mgr_cls.create_partition_from_metadata(
+                index=self.index, columns=self.columns
+            )
+
     @lazy_metadata_decorator(apply_axis="both")
     def broadcast_apply_select_indices(
         self,
@@ -3351,7 +3370,7 @@ def broadcast_apply_full_axis(
         if other is not None:
             if not isinstance(other, list):
                 other = [other]
-            other = [o._partitions for o in other] if len(other) else None
+            other = [o._extract_partitions() for o in other] if len(other) else None
 
         if apply_indices is not None:
             numeric_indices = self.get_axis(axis ^ 1).get_indexer_for(apply_indices)

modin/core/dataframe/pandas/partitioning/partition_manager.py

@@ -175,6 +175,24 @@ def preprocess_func(cls, map_func):
 
     # END Abstract Methods
 
+    @classmethod
+    def create_partition_from_metadata(cls, **metadata):
+        """
+        Create NumPy array of partitions that holds an empty dataframe with given metadata.
+
+        Parameters
+        ----------
+        **metadata : dict
+            Metadata that has to be wrapped in a partition.
+
+        Returns
+        -------
+        np.ndarray
+            A NumPy 2D array of a single partition which contains the data.
+        """
+        metadata_dataframe = pandas.DataFrame(**metadata)
+        return np.array([[cls._partition_class.put(metadata_dataframe)]])
+
     @classmethod
     def column_partitions(cls, partitions, full_axis=True):
         """
@@ -1113,6 +1131,8 @@ def combine(cls, partitions):
         np.ndarray
             A NumPy 2D array of a single partition.
         """
+        if partitions.size <= 1:
+            return partitions
 
         def to_pandas_remote(df, partition_shape, *dfs):
             """Copy of ``cls.to_pandas()`` method adapted for a remote function."""

modin/pandas/test/dataframe/test_join_sort.py

@@ -383,6 +383,13 @@ def test_merge(test_data, test_data2):
         modin_df.merge("Non-valid type")
 
 
+def test_merge_empty():
+    data = np.random.uniform(0, 100, size=(2**6, 2**6))
+    pandas_df = pandas.DataFrame(data)
+    modin_df = pd.DataFrame(data)
+    eval_general(modin_df, pandas_df, lambda df: df.merge(df.iloc[:0]))
+
+
 def test_merge_with_mi_columns():
     modin_df1, pandas_df1 = create_test_dfs(
         {