Handle nested string columns with no children in contiguous_split.

CHANGELOG.md

@@ -175,6 +175,7 @@
 - PR #6855 Fix `.str.replace_with_backrefs` docs examples
 - PR #6853 Fix contiguous split of null string columns
 - PR #6861 Fix compile error in type_dispatch_benchmark.cu
+- PR #6864 Handle contiguous_split corner case for nested string columns with no children
 
 
 # cuDF 0.16.0 (21 Oct 2020)

cpp/src/copying/contiguous_split.cu

@@ -390,10 +390,15 @@ struct buf_info_functor {
     }
 
     // info for the data buffer
-    *current = src_buf_info(
-      col.type().id(), nullptr, offset_stack_pos, parent_offset_index, false, col.offset());
+    if(col.head()){
+      *current = src_buf_info(
+        col.type().id(), nullptr, offset_stack_pos, parent_offset_index, false, col.offset());
 
-    return {current + 1, offset_stack_pos + offset_depth};
+      current++;
+      offset_stack_pos += offset_depth;
+    }
+
+    return {current, offset_stack_pos};
   }
 
  private:
@@ -419,38 +424,48 @@ std::pair<src_buf_info*, size_type> buf_info_functor::operator()<cudf::string_vi
   int parent_offset_index,
   int offset_depth)
 {
-  strings_column_view scv(col);
-
   if (col.nullable()) {
     std::tie(current, offset_stack_pos) =
       add_null_buffer(col, current, offset_stack_pos, parent_offset_index, offset_depth);
   }
 
-  auto offset_col = current;
+  // string columns can have no children and still be valid
+  if (col.num_children() > 0) {
+    strings_column_view scv(col);
 
-  // info for the offsets buffer
-  *current = src_buf_info(type_id::INT32,
-                          scv.offsets().begin<cudf::id_to_type<type_id::INT32>>(),
-                          offset_stack_pos,
-                          parent_offset_index,
-                          false,
-                          col.offset());
+    auto offset_col = current;
 
-  // prevent appending buf_info for non-exist chars buffer
-  if (scv.chars_size() > 0) {
-    current++;
-    offset_stack_pos += offset_depth;
+    // info for the offsets buffer
+    if(scv.offsets().head()){
+      CUDF_EXPECTS(scv.offsets().nullable() == false, "Encountered nullable string offsets column");
+      *current = src_buf_info(type_id::INT32,
+                              scv.offsets().begin<cudf::id_to_type<type_id::INT32>>(),
+                              offset_stack_pos,
+                              parent_offset_index,
+                              false,
+                              col.offset());
 
-    // since we are crossing an offset boundary, our offset_depth and parent_offset_index go up.
-    offset_depth++;
-    parent_offset_index = offset_col - head;
+      current++;
+      offset_stack_pos += offset_depth;
+    }
 
-    // info for the chars buffer
-    *current = src_buf_info(
-      type_id::INT8, nullptr, offset_stack_pos, parent_offset_index, false, col.offset());
+    // prevent appending buf_info for non-exist chars buffer
+    if (scv.chars().head()){
+      CUDF_EXPECTS(scv.chars().nullable() == false, "Encountered nullable string chars column");
+
+      // since we are crossing an offset boundary, our offset_depth and parent_offset_index go up.
+      offset_depth++;
+      parent_offset_index = offset_col - head;
+
+      // info for the chars buffer
+      *current = src_buf_info(
+        type_id::INT8, nullptr, offset_stack_pos, parent_offset_index, false, col.offset());
+      current++;
+      offset_stack_pos += offset_depth;
+    }
   }
 
-  return {current + 1, offset_stack_pos + offset_depth};
+  return {current, offset_stack_pos};
 }
 
 template <>

cpp/tests/copying/split_tests.cpp

@@ -16,6 +16,7 @@
 
 #include <cudf/column/column_factories.hpp>
 #include <cudf/copying.hpp>
+#include <cudf/strings/detail/utilities.hpp>
 #include <cudf/utilities/type_dispatcher.hpp>
 
 #include <cudf_test/base_fixture.hpp>
@@ -1276,6 +1277,21 @@ TEST_F(ContiguousSplitTableCornerCases, PreSplitTable)
   }
 }
 
+TEST_F(ContiguousSplitTableCornerCases, NestedEmptyStrings)
+{
+  {
+    auto empty_string = cudf::strings::detail::make_empty_strings_column();
+    auto offsets      = cudf::test::fixed_width_column_wrapper<int>({0, 1});
+    auto list         = cudf::make_lists_column(
+      1, offsets.release(), std::move(empty_string), 0, rmm::device_buffer{0});
+
+    cudf::table_view src_table({static_cast<cudf::column_view>(*list)});
+
+    std::vector<cudf::size_type> splits({0});
+    EXPECT_NO_THROW(contiguous_split(src_table, splits));
+  }
+}
+
 struct ContiguousSplitNestedTypesTest : public cudf::test::BaseFixture {
 };