FIX-modin-project#1851: Squash multiple LogicalProject nodes

In case of multiple sequential LogicalProject nodes, Calcite
squashes them, but gets the column names from the first one.
To fix the issue, squash on the Modin side and keep the
names from the last one.

Signed-off-by: Andrey Pavlenko <andrey.a.pavlenko@gmail.com>

modin/experimental/core/execution/native/implementations/hdk_on_native/calcite_builder.py

@@ -738,6 +738,18 @@ def _push(self, node):
  node : CalciteBaseNode
  A node to add.
  """
+ if (
+ len(self.res) != 0
+ and isinstance(node, CalciteProjectionNode)
+ and isinstance(self.res[-1], CalciteProjectionNode)
+ and all(isinstance(expr, CalciteInputRefExpr) for expr in node.exprs)
+ ):
+ # Replace the last CalciteProjectionNode with this one and
+ # translate the input refs.
+ exprs = self.res.pop().exprs
+ node = CalciteProjectionNode(
+ node.fields, [exprs[expr.input] for expr in node.exprs]
+ )
  self.res.append(node)
 
  def _last(self):

modin/experimental/core/execution/native/implementations/hdk_on_native/dataframe/dataframe.py

@@ -2041,9 +2041,7 @@ def _materialize(self):
  new_table, unsupported_cols=[], encode_col_names=False
  )[0]
  else:
- partitions = self._partition_mgr_cls.run_exec_plan(
- self._op, self._table_cols
- )
+ partitions = self._partition_mgr_cls.run_exec_plan(self._op)
 
  self._partitions = partitions
  self._op = FrameNode(self)

modin/experimental/core/execution/native/implementations/hdk_on_native/dataframe/utils.py

@@ -429,12 +429,12 @@ def to_empty_pandas_df(df):
  index_cols = None
  else:
  index_cols = ColNameCodec.mangle_index_names(merged.index.names)
- for orig_name, mangled_name in zip(merged.index.names, index_cols):
+ for name in index_cols:
  # Using _dtypes here since it contains all column names,
  # including the index.
- df = left if mangled_name in left._dtypes else right
- exprs[orig_name] = df.ref(mangled_name)
- new_dtypes.append(df._dtypes[mangled_name])
+ df = left if name in left._dtypes else right
+ exprs[name] = df.ref(name)
+ new_dtypes.append(df._dtypes[name])
 
  left_col_names = set(left.columns)
  right_col_names = set(right.columns)

modin/experimental/core/execution/native/implementations/hdk_on_native/partitioning/partition_manager.py

@@ -227,16 +227,14 @@ def is_supported_dtype(dtype):
  )
 
  @classmethod
- def run_exec_plan(cls, plan, columns):
+ def run_exec_plan(cls, plan):
  """
  Run execution plan in HDK storage format to materialize frame.
 
  Parameters
  ----------
  plan : DFAlgNode
  A root of an execution plan tree.
- columns : list of str
- A frame column names.
 
  Returns
  -------
@@ -256,10 +254,6 @@ def run_exec_plan(cls, plan, columns):
  calcite_json = "execute calcite " + calcite_json
  table = worker.executeRA(calcite_json)
 
- # workaround for https://github.com/modin-project/modin/issues/1851
- if DoUseCalcite.get():
- table._column_names = [ColNameCodec.encode(c) for c in columns]
-
  res = np.empty((1, 1), dtype=np.dtype(object))
  res[0][0] = cls._partition_class(table)