Refactor Python factories and remove usage of Table for libcudf output handling

python/cudf/cudf/_lib/avro.pyx

@@ -12,6 +12,7 @@ from cudf._lib.cpp.io.types cimport table_with_metadata
 from cudf._lib.cpp.types cimport size_type
 from cudf._lib.io.utils cimport make_source_info
 from cudf._lib.table cimport Table
+from cudf._lib.utils cimport data_from_unique_ptr
 
 
 cpdef read_avro(datasource, columns=None, skip_rows=-1, num_rows=-1):
@@ -52,4 +53,4 @@ cpdef read_avro(datasource, columns=None, skip_rows=-1, num_rows=-1):
 
     names = [name.decode() for name in c_result.metadata.column_names]
 
-    return Table.from_unique_ptr(move(c_result.tbl), column_names=names)
+    return data_from_unique_ptr(move(c_result.tbl), column_names=names)

python/cudf/cudf/_lib/concat.pyx

@@ -15,6 +15,7 @@ from cudf._lib.cpp.concatenate cimport (
 from cudf._lib.cpp.table.table cimport table, table_view
 from cudf._lib.table cimport Table
 from cudf._lib.utils cimport (
+    data_from_unique_ptr,
     make_column_views,
     make_table_data_views,
     make_table_views,
@@ -52,7 +53,8 @@ cpdef concat_tables(object tables, bool ignore_index=False):
         c_views = make_table_data_views(tables)
     with nogil:
         c_result = move(libcudf_concatenate_tables(c_views))
-    return Table.from_unique_ptr(
+
+    return data_from_unique_ptr(
         move(c_result),
         column_names=tables[0]._column_names,
         index_names=None if ignore_index else tables[0]._index_names

python/cudf/cudf/_lib/copying.pyx

@@ -11,6 +11,7 @@ from libcpp.utility cimport move
 from libcpp.vector cimport vector
 
 from rmm._lib.device_buffer cimport DeviceBuffer
+
 from cudf.core.buffer import Buffer
 
 from cudf._lib.column cimport Column
@@ -35,6 +36,7 @@ from cudf._lib.cpp.scalar.scalar cimport scalar
 from cudf._lib.cpp.table.table cimport table
 from cudf._lib.cpp.table.table_view cimport table_view
 from cudf._lib.cpp.types cimport size_type
+from cudf._lib.utils cimport data_from_table_view, data_from_unique_ptr
 
 # workaround for https://github.com/cython/cython/issues/3885
 ctypedef const scalar constscalar
@@ -178,7 +180,7 @@ def gather(
             )
         )
 
-    return Table.from_unique_ptr(
+    return data_from_unique_ptr(
         move(c_result),
         column_names=source_table._column_names,
         index_names=(
@@ -210,19 +212,17 @@ def _scatter_table(Table source_table, Column scatter_map,
             )
         )
 
-    out_table = Table.from_unique_ptr(
+    data, _ = data_from_unique_ptr(
         move(c_result),
         column_names=target_table._column_names,
         index_names=None
     )
 
-    out_table._index = (
+    return data, (
         None if target_table._index is None else target_table._index.copy(
             deep=False)
     )
 
-    return out_table
-
 
 def _scatter_scalar(scalars, Column scatter_map,
                     Table target_table, bool bounds_check=True):
@@ -250,19 +250,17 @@ def _scatter_scalar(scalars, Column scatter_map,
             )
         )
 
-    out_table = Table.from_unique_ptr(
+    data, _ = data_from_unique_ptr(
         move(c_result),
         column_names=target_table._column_names,
         index_names=None
     )
 
-    out_table._index = (
+    return data, (
         None if target_table._index is None else target_table._index.copy(
             deep=False)
     )
 
-    return out_table
-
 
 def scatter(object input, object scatter_map, Table target,
             bool bounds_check=True):
@@ -306,7 +304,7 @@ def _reverse_table(Table source_table):
             reverse_table_view
         ))
 
-    return Table.from_unique_ptr(
+    return data_from_unique_ptr(
         move(c_result),
         column_names=source_table._column_names,
         index_names=source_table._index_names
@@ -371,7 +369,7 @@ def table_empty_like(Table input_table, bool keep_index=True):
     with nogil:
         c_result = move(cpp_copying.empty_like(input_table_view))
 
-    return Table.from_unique_ptr(
+    return data_from_unique_ptr(
         move(c_result),
         column_names=input_table._column_names,
         index_names=(
@@ -434,8 +432,8 @@ def table_slice(Table input_table, object indices, bool keep_index=True):
         )
 
     num_of_result_cols = c_result.size()
-    result =[
-        Table.from_table_view(
+    return [
+        data_from_table_view(
             c_result[i],
             input_table,
             column_names=input_table._column_names,
@@ -446,8 +444,6 @@ def table_slice(Table input_table, object indices, bool keep_index=True):
             )
         ) for i in range(num_of_result_cols)]
 
-    return result
-
 
 def column_split(Column input_column, object splits):
 
@@ -505,8 +501,8 @@ def table_split(Table input_table, object splits, bool keep_index=True):
         )
 
     num_of_result_cols = c_result.size()
-    result = [
-        Table.from_table_view(
+    return [
+        data_from_table_view(
             c_result[i],
             input_table,
             column_names=input_table._column_names,
@@ -515,8 +511,6 @@ def table_split(Table input_table, object splits, bool keep_index=True):
             else None
         ) for i in range(num_of_result_cols)]
 
-    return result
-
 
 def _copy_if_else_column_column(Column lhs, Column rhs, Column boolean_mask):
 
@@ -642,7 +636,7 @@ def _boolean_mask_scatter_table(Table input_table, Table target_table,
             )
         )
 
-    return Table.from_unique_ptr(
+    return data_from_unique_ptr(
         move(c_result),
         column_names=target_table._column_names,
         index_names=target_table._index._column_names
@@ -672,13 +666,15 @@ def _boolean_mask_scatter_scalar(list input_scalars, Table target_table,
             )
         )
 
-    return Table.from_unique_ptr(
+    return data_from_unique_ptr(
         move(c_result),
         column_names=target_table._column_names,
         index_names=target_table._index._column_names
     )
 
 
+# TODO: This function is currently unused but should be used in
+# ColumnBase.__setitem__, see https://github.com/rapidsai/cudf/issues/8667.
 def boolean_mask_scatter(object input, Table target_table,
                          Column boolean_mask):
 
@@ -755,7 +751,7 @@ def sample(Table input, size_type n,
             cpp_copying.sample(tbl_view, n, replacement, seed)
         )
 
-    return Table.from_unique_ptr(
+    return data_from_unique_ptr(
         move(c_output),
         column_names=input._column_names,
         index_names=(
@@ -884,12 +880,12 @@ cdef class _CPackedColumns:
         return p
 
     def unpack(self):
-        output_table = Table.from_table_view(
+        output_table = Table(*data_from_table_view(
             cpp_copying.unpack(self.c_obj),
             self,
             self.column_names,
             self.index_names
-        )
+        ))
 
         for name, dtype in self.column_dtypes.items():
             output_table._data[name] = (

python/cudf/cudf/_lib/csv.pyx

@@ -40,6 +40,7 @@ from cudf._lib.cpp.io.types cimport (
 from cudf._lib.cpp.table.table_view cimport table_view
 from cudf._lib.io.utils cimport make_sink_info, make_source_info
 from cudf._lib.table cimport Table, make_table_view
+from cudf._lib.utils cimport data_from_unique_ptr
 
 ctypedef int32_t underlying_type_t_compression
 
@@ -393,7 +394,7 @@ def read_csv(
         c_result = move(cpp_read_csv(read_csv_options_c))
 
     meta_names = [name.decode() for name in c_result.metadata.column_names]
-    df = cudf.DataFrame._from_table(Table.from_unique_ptr(
+    df = cudf.DataFrame._from_data(*data_from_unique_ptr(
         move(c_result.tbl),
         column_names=meta_names
     ))

python/cudf/cudf/_lib/filling.pyx

@@ -16,6 +16,7 @@ from cudf._lib.cpp.table.table_view cimport table_view
 from cudf._lib.cpp.types cimport size_type
 from cudf._lib.scalar cimport DeviceScalar
 from cudf._lib.table cimport Table
+from cudf._lib.utils cimport data_from_unique_ptr
 
 
 def fill_in_place(Column destination, int begin, int end, DeviceScalar value):
@@ -70,7 +71,7 @@ def _repeat_via_column(Table inp, Column count, bool check_count):
             c_check_count
         ))
 
-    return Table.from_unique_ptr(
+    return data_from_unique_ptr(
         move(c_result),
         column_names=inp._column_names,
         index_names=inp._index_names
@@ -87,7 +88,7 @@ def _repeat_via_size_type(Table inp, size_type count):
             count
         ))
 
-    return Table.from_unique_ptr(
+    return data_from_unique_ptr(
         move(c_result),
         column_names=inp._column_names,
         index_names=inp._index_names

python/cudf/cudf/_lib/groupby.pyx

@@ -22,6 +22,8 @@ from libcpp.pair cimport pair
 from libcpp.utility cimport move
 from libcpp.vector cimport vector
 
+import cudf
+
 from cudf._lib.column cimport Column
 from cudf._lib.scalar cimport DeviceScalar
 from cudf._lib.table cimport Table
@@ -39,6 +41,7 @@ from cudf._lib.cpp.scalar.scalar cimport scalar
 from cudf._lib.cpp.table.table cimport table, table_view
 from cudf._lib.cpp.types cimport size_type
 from cudf._lib.cpp.utilities.host_span cimport host_span
+from cudf._lib.utils cimport data_from_unique_ptr
 
 # The sets below define the possible aggregations that can be performed on
 # different dtypes. These strings must be elements of the AggregationKind enum.
@@ -91,11 +94,11 @@ cdef class GroupBy:
         c_grouped_values = move(c_groups.values)
         c_group_offsets = c_groups.offsets
 
-        grouped_keys = Table.from_unique_ptr(
+        grouped_keys = cudf.Index._from_data(*data_from_unique_ptr(
             move(c_grouped_keys),
             column_names=range(c_grouped_keys.get()[0].num_columns())
-        )
-        grouped_values = Table.from_unique_ptr(
+        ))
+        grouped_values = data_from_unique_ptr(
             move(c_grouped_values),
             index_names=values._index_names,
             column_names=values._column_names
@@ -197,7 +200,7 @@ cdef class GroupBy:
             else:
                 raise
 
-        grouped_keys = Table.from_unique_ptr(
+        grouped_keys, _ = data_from_unique_ptr(
             move(c_result.first),
             column_names=self.keys._column_names
         )
@@ -213,7 +216,7 @@ cdef class GroupBy:
                     Column.from_unique_ptr(move(c_result.second[i].results[j]))
                 )
 
-        return Table(data=result_data, index=grouped_keys)
+        return result_data, cudf.Index._from_data(grouped_keys)
 
     def shift(self, Table values, int periods, list fill_values):
         cdef table_view view = values.view()
@@ -238,16 +241,16 @@ cdef class GroupBy:
                 self.c_obj.get()[0].shift(view, offsets, c_fill_values)
             )
 
-        grouped_keys = Table.from_unique_ptr(
+        grouped_keys = cudf.Index._from_data(*data_from_unique_ptr(
             move(c_result.first),
             column_names=self.keys._column_names
-        )
+        ))
 
-        shifted = Table.from_unique_ptr(
+        shifted, _ = data_from_unique_ptr(
             move(c_result.second), column_names=values._column_names
         )
 
-        return Table(data=shifted._data, index=grouped_keys)
+        return shifted, grouped_keys
 
     def replace_nulls(self, Table values, object method):
         cdef table_view val_view = values.view()
@@ -265,12 +268,10 @@ cdef class GroupBy:
                 self.c_obj.get()[0].replace_nulls(val_view, policies)
             )
 
-        grouped_result = Table.from_unique_ptr(
+        return data_from_unique_ptr(
             move(c_result.second), column_names=values._column_names
-        )
+        )[0]
 
-        result = Table(data=grouped_result._data)
-        return result
 
 _GROUPBY_SCANS = {"cumcount", "cumsum", "cummin", "cummax"}
 

python/cudf/cudf/_lib/hash.pyx

@@ -15,6 +15,7 @@ from cudf._lib.cpp.partitioning cimport hash_partition as cpp_hash_partition
 from cudf._lib.cpp.table.table cimport table
 from cudf._lib.cpp.table.table_view cimport table_view
 from cudf._lib.table cimport Table
+from cudf._lib.utils cimport data_from_unique_ptr
 
 
 def hash_partition(Table source_table, object columns_to_hash,
@@ -41,12 +42,14 @@ def hash_partition(Table source_table, object columns_to_hash,
     # the original table (`source_table`) is empty. We need to
     # return a list of zeros in this case.
     return (
-        Table.from_unique_ptr(
+        *data_from_unique_ptr(
             move(c_result.first),
             column_names=source_table._column_names,
-            index_names=source_table._index_names if(
-                keep_index is True)
-            else None
+            index_names=(
+                source_table._index_names
+                if keep_index is True
+                else None
+            )
 
         ),
         list(c_result.second) if c_result.second.size()

python/cudf/cudf/_lib/interop.pyx

@@ -21,6 +21,7 @@ from cudf._lib.cpp.interop cimport (
 from cudf._lib.cpp.table.table cimport table
 from cudf._lib.cpp.table.table_view cimport table_view
 from cudf._lib.table cimport Table
+from cudf._lib.utils cimport data_from_unique_ptr
 
 
 def from_dlpack(dlpack_capsule):
@@ -40,7 +41,7 @@ def from_dlpack(dlpack_capsule):
             cpp_from_dlpack(dlpack_tensor)
         )
 
-    res = Table.from_unique_ptr(
+    res = data_from_unique_ptr(
         move(c_result),
         column_names=range(0, c_result.get()[0].num_columns())
     )
@@ -164,10 +165,8 @@ def from_arrow(
     with nogil:
         c_result = move(cpp_from_arrow(cpp_arrow_table.get()[0]))
 
-    out_table = Table.from_unique_ptr(
+    return data_from_unique_ptr(
         move(c_result),
         column_names=column_names,
         index_names=index_names
     )
-
-    return out_table