From a00295a4db71c7c0c1396d2c0539eecee74a4ac5 Mon Sep 17 00:00:00 2001
From: Karthikeyan <6488848+karthikeyann@users.noreply.github.com>
Date: Fri, 17 Feb 2023 01:37:56 +0530
Subject: [PATCH] Fix bug in all-null list due to join_list_elements special
 handling (#12767)

All-null list is considered as empty list in `join_list_elements`. So, nulls of children are replaced by `narep` before passing to `join_list_elements` API.
Related https://github.com/rapidsai/cudf/issues/12766

Authors:
  - Karthikeyan (https://github.com/karthikeyann)

Approvers:
  - David Wendt (https://github.com/davidwendt)
  - Nghia Truong (https://github.com/ttnghia)
  - Bradley Dice (https://github.com/bdice)

URL: https://github.com/rapidsai/cudf/pull/12767
---
 cpp/src/io/json/write_json.cu |  8 ++++++--
 cpp/tests/io/json_writer.cpp  | 32 ++++++++++++++++++++++++++++++++
 2 files changed, 38 insertions(+), 2 deletions(-)

diff --git a/cpp/src/io/json/write_json.cu b/cpp/src/io/json/write_json.cu
index 9849629015d..a7ae4d3bdd1 100644
--- a/cpp/src/io/json/write_json.cu
+++ b/cpp/src/io/json/write_json.cu
@@ -417,7 +417,9 @@ struct column_to_strings_fn {
     auto child_view            = lists_column_view(column).get_sliced_child(stream_);
     auto constexpr child_index = lists_column_view::child_column_index;
     auto list_string           = [&]() {
-      auto child_string = [&]() {
+      // nulls are replaced due to special handling of all-null lists as empty lists
+      // by join_list_elements
+      auto child_string_with_null = [&]() {
         if (child_view.type().id() == type_id::STRUCT) {
           return (*this).template operator()<cudf::struct_view>(
             child_view,
@@ -431,7 +433,9 @@ struct column_to_strings_fn {
         } else {
           return cudf::type_dispatcher(child_view.type(), *this, child_view);
         }
-      }();
+      };
+      auto child_string = cudf::strings::detail::replace_nulls(
+        child_string_with_null()->view(), narep, stream_, rmm::mr::get_current_device_resource());
       auto const list_child_string =
         column_view(column.type(),
                     column.size(),
diff --git a/cpp/tests/io/json_writer.cpp b/cpp/tests/io/json_writer.cpp
index d129ed306e4..702315d6a97 100644
--- a/cpp/tests/io/json_writer.cpp
+++ b/cpp/tests/io/json_writer.cpp
@@ -362,4 +362,36 @@ TEST_F(JsonWriterTest, SpecialChars)
   EXPECT_EQ(expected, output_string);
 }
 
+TEST_F(JsonWriterTest, NullList)
+{
+  std::string const data = R"(
+{"a": [null], "b": [[1, 2, 3], [null], [null, null, null], [4, null, 5]]}
+{"a": [2, null, null, 3] , "b": null}
+{"a": [null, null, 4], "b": [[2, null], null]}
+{"a": [5, null, null], "b": [null, [3, 4, 5]]} )";
+  cudf::io::json_reader_options in_options =
+    cudf::io::json_reader_options::builder(cudf::io::source_info{data.data(), data.size()})
+      .lines(true);
+
+  cudf::io::table_with_metadata result = cudf::io::read_json(in_options);
+  cudf::table_view tbl_view            = result.tbl->view();
+  cudf::io::table_metadata mt{result.metadata};
+
+  std::vector<char> out_buffer;
+  auto destination     = cudf::io::sink_info(&out_buffer);
+  auto options_builder = cudf::io::json_writer_options_builder(destination, tbl_view)
+                           .include_nulls(true)
+                           .metadata(mt)
+                           .lines(true)
+                           .na_rep("null");
+
+  cudf::io::write_json(options_builder.build(), rmm::mr::get_current_device_resource());
+  std::string const expected = R"({"a":[null],"b":[[1,2,3],[null],[null,null,null],[4,null,5]]}
+{"a":[2,null,null,3],"b":null}
+{"a":[null,null,4],"b":[[2,null],null]}
+{"a":[5,null,null],"b":[null,[3,4,5]]}
+)";
+  EXPECT_EQ(expected, std::string(out_buffer.data(), out_buffer.size()));
+}
+
 CUDF_TEST_PROGRAM_MAIN()