Remove Cython APIs for table view generation (#9199)

This PR removes the Table methods used to generate `cudf::table_view` objects in Cython and switches all code paths to use the underlying `cdef` function that those functions originally called under the hood. Many of the view-generating functions were actually never used (all the mutable view generators and `index_view`), so those have been removed. Moving in this direction has two benefits. First, it allows us to change our Cython APIs to accept iterables of columns rather than `Table` objects, which in turn means that we can get rid of various Column->Table->Column round-trips that are currently in place just to call Cython APIs on Columns that are only defined on Tables. Second, the Table class is now pure Python, so its logic can be merged into the Frame class.

Authors:
  - Vyas Ramasubramani (https://github.com/vyasr)

Approvers:
  - Michael Wang (https://github.com/isVoid)
  - GALI PREM SAGAR (https://github.com/galipremsagar)

URL: #9199

python/cudf/cudf/_lib/concat.pyx

@@ -13,13 +13,8 @@ from cudf._lib.cpp.concatenate cimport (
     concatenate_tables as libcudf_concatenate_tables,
 )
 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,
-)
+from cudf._lib.table cimport Table, table_view_from_table
+from cudf._lib.utils cimport data_from_unique_ptr, make_column_views
 
 from cudf.core.buffer import Buffer
 
@@ -47,10 +42,9 @@ cpdef concat_columns(object columns):
 cpdef concat_tables(object tables, bool ignore_index=False):
     cdef unique_ptr[table] c_result
     cdef vector[table_view] c_views
-    if ignore_index is False:
-        c_views = make_table_views(tables)
-    else:
-        c_views = make_table_data_views(tables)
+    c_views.reserve(len(tables))
+    for tbl in tables:
+        c_views.push_back(table_view_from_table(tbl, ignore_index))
     with nogil:
         c_result = move(libcudf_concatenate_tables(c_views))
 

python/cudf/cudf/_lib/copying.pyx

@@ -19,7 +19,11 @@ from cudf._lib.column cimport Column
 from cudf._lib.scalar import as_device_scalar
 
 from cudf._lib.scalar cimport DeviceScalar
-from cudf._lib.table cimport Table, make_table_view
+from cudf._lib.table cimport (
+    Table,
+    table_view_from_columns,
+    table_view_from_table,
+)
 
 from cudf._lib.reduce import minmax
 from cudf.core.abc import Serializable
@@ -160,11 +164,9 @@ def gather(
                              f" rows.")
 
     cdef unique_ptr[table] c_result
-    cdef table_view source_table_view
-    if keep_index is True:
-        source_table_view = source_table.view()
-    else:
-        source_table_view = source_table.data_view()
+    cdef table_view source_table_view = table_view_from_table(
+        source_table, not keep_index
+    )
     cdef column_view gather_map_view = gather_map.view()
     cdef cpp_copying.out_of_bounds_policy policy = (
         cpp_copying.out_of_bounds_policy.NULLIFY if nullify
@@ -200,7 +202,8 @@ def scatter(object source, Column scatter_map, Column target_column,
     """
 
     cdef column_view scatter_map_view = scatter_map.view()
-    cdef table_view target_table_view = make_table_view((target_column,))
+    cdef table_view target_table_view = table_view_from_columns(
+        (target_column,))
     cdef bool c_bounds_check = bounds_check
     cdef unique_ptr[table] c_result
 
@@ -212,7 +215,7 @@ def scatter(object source, Column scatter_map, Column target_column,
     cdef DeviceScalar slr
 
     if isinstance(source, Column):
-        source_table_view = make_table_view((<Column> source,))
+        source_table_view = table_view_from_columns((<Column> source,))
 
         with nogil:
             c_result = move(
@@ -263,7 +266,7 @@ def _reverse_column(Column source_column):
 
 
 def _reverse_table(Table source_table):
-    cdef table_view reverse_table_view = source_table.view()
+    cdef table_view reverse_table_view = table_view_from_columns(source_table)
 
     cdef unique_ptr[table] c_result
     with nogil:
@@ -325,11 +328,9 @@ def column_allocate_like(Column input_column, size=None):
 
 def table_empty_like(Table input_table, bool keep_index=True):
 
-    cdef table_view input_table_view
-    if keep_index is True:
-        input_table_view = input_table.view()
-    else:
-        input_table_view = input_table.data_view()
+    cdef table_view input_table_view = table_view_from_table(
+        input_table, not keep_index
+    )
 
     cdef unique_ptr[table] c_result
 
@@ -376,11 +377,9 @@ def column_slice(Column input_column, object indices):
 
 def table_slice(Table input_table, object indices, bool keep_index=True):
 
-    cdef table_view input_table_view
-    if keep_index is True:
-        input_table_view = input_table.view()
-    else:
-        input_table_view = input_table.data_view()
+    cdef table_view input_table_view = table_view_from_table(
+        input_table, not keep_index
+    )
 
     cdef vector[size_type] c_indices
     c_indices.reserve(len(indices))
@@ -445,11 +444,9 @@ def column_split(Column input_column, object splits):
 
 def table_split(Table input_table, object splits, bool keep_index=True):
 
-    cdef table_view input_table_view
-    if keep_index is True:
-        input_table_view = input_table.view()
-    else:
-        input_table_view = input_table.data_view()
+    cdef table_view input_table_view = table_view_from_table(
+        input_table, not keep_index
+    )
 
     cdef vector[size_type] c_splits
     c_splits.reserve(len(splits))
@@ -588,8 +585,8 @@ def copy_if_else(object lhs, object rhs, Column boolean_mask):
 def _boolean_mask_scatter_table(Table input_table, Table target_table,
                                 Column boolean_mask):
 
-    cdef table_view input_table_view = input_table.view()
-    cdef table_view target_table_view = target_table.view()
+    cdef table_view input_table_view = table_view_from_columns(input_table)
+    cdef table_view target_table_view = table_view_from_columns(target_table)
     cdef column_view boolean_mask_view = boolean_mask.view()
 
     cdef unique_ptr[table] c_result
@@ -619,7 +616,7 @@ def _boolean_mask_scatter_scalar(list input_scalars, Table target_table,
     for scl in input_scalars:
         input_scalar_vector.push_back(reference_wrapper[constscalar](
             scl.get_raw_ptr()[0]))
-    cdef table_view target_table_view = target_table.view()
+    cdef table_view target_table_view = table_view_from_columns(target_table)
     cdef column_view boolean_mask_view = boolean_mask.view()
 
     cdef unique_ptr[table] c_result
@@ -702,9 +699,7 @@ def get_element(Column input_column, size_type index):
 
 def sample(Table input, size_type n,
            bool replace, int64_t seed, bool keep_index=True):
-    cdef table_view tbl_view = (
-        input.view() if keep_index else input.data_view()
-    )
+    cdef table_view tbl_view = table_view_from_table(input, not keep_index)
     cdef cpp_copying.sample_with_replacement replacement
 
     if replace:
@@ -764,10 +759,11 @@ cdef class _CPackedColumns:
             or input_table.index.stop != len(input_table)
             or input_table.index.step != 1
         ):
-            input_table_view = input_table.view()
+            input_table_view = table_view_from_table(input_table)
             p.index_names = input_table._index_names
         else:
-            input_table_view = input_table.data_view()
+            input_table_view = table_view_from_table(
+                input_table, ignore_index=True)
 
         p.column_names = input_table._column_names
         p.column_dtypes = {}

python/cudf/cudf/_lib/csv.pyx

@@ -44,7 +44,11 @@ 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.table cimport (
+    Table,
+    table_view_from_columns,
+    table_view_from_table,
+)
 from cudf._lib.utils cimport data_from_unique_ptr
 
 ctypedef int32_t underlying_type_t_compression
@@ -458,8 +462,9 @@ cpdef write_csv(
     --------
     cudf.to_csv
     """
-    cdef table_view input_table_view = \
-        table.view() if index is True else table.data_view()
+    cdef table_view input_table_view = table_view_from_table(
+        table, not index
+    )
     cdef bool include_header_c = header
     cdef char delim_c = ord(sep)
     cdef string line_term_c = line_terminator.encode()

python/cudf/cudf/_lib/filling.pyx

@@ -15,7 +15,7 @@ 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.scalar cimport DeviceScalar
-from cudf._lib.table cimport Table
+from cudf._lib.table cimport Table, table_view_from_table
 from cudf._lib.utils cimport data_from_unique_ptr
 
 
@@ -59,7 +59,7 @@ def repeat(Table inp, object count, bool check_count=False):
 
 
 def _repeat_via_column(Table inp, Column count, bool check_count):
-    cdef table_view c_inp = inp.view()
+    cdef table_view c_inp = table_view_from_table(inp)
     cdef column_view c_count = count.view()
     cdef bool c_check_count = check_count
     cdef unique_ptr[table] c_result
@@ -79,7 +79,7 @@ def _repeat_via_column(Table inp, Column count, bool check_count):
 
 
 def _repeat_via_size_type(Table inp, size_type count):
-    cdef table_view c_inp = inp.view()
+    cdef table_view c_inp = table_view_from_table(inp)
     cdef unique_ptr[table] c_result
 
     with nogil:

python/cudf/cudf/_lib/groupby.pyx

@@ -26,7 +26,7 @@ import cudf
 
 from cudf._lib.column cimport Column
 from cudf._lib.scalar cimport DeviceScalar
-from cudf._lib.table cimport Table
+from cudf._lib.table cimport Table, table_view_from_table
 
 from cudf._lib.scalar import as_device_scalar
 
@@ -74,7 +74,7 @@ cdef class GroupBy:
         else:
             c_null_handling = libcudf_types.null_policy.INCLUDE
 
-        cdef table_view keys_view = keys.view()
+        cdef table_view keys_view = table_view_from_table(keys)
 
         with nogil:
             self.c_obj.reset(
@@ -90,7 +90,7 @@ cdef class GroupBy:
 
     def groups(self, Table values):
 
-        cdef table_view values_view = values.view()
+        cdef table_view values_view = table_view_from_table(values)
 
         with nogil:
             c_groups = move(self.c_obj.get()[0].get_groups(values_view))
@@ -293,7 +293,7 @@ cdef class GroupBy:
         return self.aggregate_internal(values, aggregations)
 
     def shift(self, Table values, int periods, list fill_values):
-        cdef table_view view = values.view()
+        cdef table_view view = table_view_from_table(values)
         cdef size_type num_col = view.num_columns()
         cdef vector[size_type] offsets = vector[size_type](num_col, periods)
 
@@ -329,7 +329,7 @@ cdef class GroupBy:
         return shifted, grouped_keys
 
     def replace_nulls(self, Table values, object method):
-        cdef table_view val_view = values.view()
+        cdef table_view val_view = table_view_from_table(values)
         cdef pair[unique_ptr[table], unique_ptr[table]] c_result
         cdef replace_policy policy = (
             replace_policy.PRECEDING

python/cudf/cudf/_lib/hash.pyx

@@ -14,19 +14,17 @@ from cudf._lib.cpp.hash cimport hash as cpp_hash
 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.table cimport Table, table_view_from_table
 from cudf._lib.utils cimport data_from_unique_ptr
 
 
 def hash_partition(Table source_table, object columns_to_hash,
                    int num_partitions, bool keep_index=True):
     cdef vector[libcudf_types.size_type] c_columns_to_hash = columns_to_hash
     cdef int c_num_partitions = num_partitions
-    cdef table_view c_source_view
-    if keep_index is True:
-        c_source_view = source_table.view()
-    else:
-        c_source_view = source_table.data_view()
+    cdef table_view c_source_view = table_view_from_table(
+        source_table, not keep_index
+    )
 
     cdef pair[unique_ptr[table], vector[libcudf_types.size_type]] c_result
     with nogil:
@@ -59,7 +57,8 @@ def hash_partition(Table source_table, object columns_to_hash,
 
 def hash(Table source_table, object initial_hash_values=None, int seed=0):
     cdef vector[uint32_t] c_initial_hash = initial_hash_values or []
-    cdef table_view c_source_view = source_table.data_view()
+    cdef table_view c_source_view = table_view_from_table(
+        source_table, ignore_index=True)
 
     cdef unique_ptr[column] c_result
     with nogil:

python/cudf/cudf/_lib/interop.pyx

@@ -20,7 +20,7 @@ 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.table cimport Table, table_view_from_table
 from cudf._lib.utils cimport data_from_unique_ptr
 
 
@@ -63,7 +63,9 @@ def to_dlpack(Table source_table):
             )
 
     cdef DLManagedTensor *dlpack_tensor
-    cdef table_view source_table_view = source_table.data_view()
+    cdef table_view source_table_view = table_view_from_table(
+        source_table, ignore_index=True
+    )
 
     with nogil:
         dlpack_tensor = cpp_to_dlpack(
@@ -128,7 +130,7 @@ def to_arrow(Table input_table,
 
     cdef vector[column_metadata] cpp_metadata = gather_metadata(metadata)
     cdef table_view input_table_view = (
-        input_table.view() if keep_index else input_table.data_view()
+        table_view_from_table(input_table, not keep_index)
     )
 
     cdef shared_ptr[CTable] cpp_arrow_table

python/cudf/cudf/_lib/join.pyx

@@ -16,15 +16,15 @@ from cudf._lib.cpp.column.column cimport column
 from cudf._lib.cpp.table.table cimport table
 from cudf._lib.cpp.table.table_view cimport table_view
 from cudf._lib.cpp.types cimport data_type, size_type, type_id
-from cudf._lib.table cimport Table, columns_from_ptr
+from cudf._lib.table cimport Table, table_view_from_table
 
 # The functions below return the *gathermaps* that represent
 # the join result when joining on the keys `lhs` and `rhs`.
 
 cpdef join(Table lhs, Table rhs, how=None):
     cdef pair[cpp_join.gather_map_type, cpp_join.gather_map_type] c_result
-    cdef table_view c_lhs = lhs.view()
-    cdef table_view c_rhs = rhs.view()
+    cdef table_view c_lhs = table_view_from_table(lhs)
+    cdef table_view c_rhs = table_view_from_table(rhs)
 
     if how == "inner":
         c_result = move(cpp_join.inner_join(
@@ -52,8 +52,8 @@ cpdef join(Table lhs, Table rhs, how=None):
 cpdef semi_join(Table lhs, Table rhs, how=None):
     # left-semi and left-anti joins
     cdef cpp_join.gather_map_type c_result
-    cdef table_view c_lhs = lhs.view()
-    cdef table_view c_rhs = rhs.view()
+    cdef table_view c_lhs = table_view_from_table(lhs)
+    cdef table_view c_rhs = table_view_from_table(rhs)
 
     if how == "leftsemi":
         c_result = move(cpp_join.left_semi_join(

python/cudf/cudf/_lib/lists.pyx

@@ -33,7 +33,7 @@ from cudf._lib.cpp.types cimport (
     size_type,
 )
 from cudf._lib.scalar cimport DeviceScalar
-from cudf._lib.table cimport Table
+from cudf._lib.table cimport Table, table_view_from_table
 from cudf._lib.types cimport (
     underlying_type_t_null_order,
     underlying_type_t_order,
@@ -63,9 +63,7 @@ def count_elements(Column col):
 
 
 def explode_outer(Table tbl, int explode_column_idx, bool ignore_index=False):
-    cdef table_view c_table_view = (
-        tbl.data_view() if ignore_index else tbl.view()
-    )
+    cdef table_view c_table_view = table_view_from_table(tbl, ignore_index)
     cdef size_type c_explode_column_idx = explode_column_idx
 
     cdef unique_ptr[table] c_result
@@ -168,7 +166,7 @@ def contains_scalar(Column col, object py_search_key):
 def concatenate_rows(Table tbl):
     cdef unique_ptr[column] c_result
 
-    cdef table_view c_table_view = tbl.view()
+    cdef table_view c_table_view = table_view_from_table(tbl)
 
     with nogil:
         c_result = move(cpp_concatenate_rows(