Fix missing trailing comma in json writer (#12688)

Fix missing trailing comma in json writer for non-Lines json format.
Updates default rows_per_chunk because `8` is too small and writer is slow.
closes #12687

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

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

URL: #12688

cpp/src/io/json/write_json.cu

@@ -581,14 +581,13 @@ std::unique_ptr<column> make_column_names_column(host_span<column_name_info cons
 
 void write_chunked(data_sink* out_sink,
                    strings_column_view const& str_column_view,
-                   std::string const& line_terminator,
+                   string_scalar const& d_line_terminator,
                    json_writer_options const& options,
                    rmm::cuda_stream_view stream,
                    rmm::mr::device_memory_resource* mr)
 {
   CUDF_EXPECTS(str_column_view.size() > 0, "Unexpected empty strings column.");
 
-  string_scalar d_line_terminator{line_terminator};
   auto p_str_col_w_nl = cudf::strings::detail::join_strings(str_column_view,
                                                             d_line_terminator,
                                                             string_scalar("", false),
@@ -609,15 +608,6 @@ void write_chunked(data_sink* out_sink,
 
     out_sink->host_write(h_bytes.data(), total_num_bytes);
   }
-
-  // Needs newline at the end, to separate from next chunk
-  if (options.is_enabled_lines()) {
-    if (out_sink->is_device_write_preferred(d_line_terminator.size())) {
-      out_sink->device_write(d_line_terminator.data(), d_line_terminator.size(), stream);
-    } else {
-      out_sink->host_write(line_terminator.data(), line_terminator.size());
-    }
-  }
 }
 
 void write_json(data_sink* out_sink,
@@ -697,7 +687,16 @@ void write_json(data_sink* out_sink,
       // struct converter for the table
       auto str_concat_col = converter(sub_view.begin(), sub_view.end(), user_column_names);
 
-      write_chunked(out_sink, str_concat_col->view(), line_terminator, options, stream, mr);
+      write_chunked(out_sink, str_concat_col->view(), d_line_terminator, options, stream, mr);
+
+      // Needs line_terminator at the end, to separate from next chunk
+      if (&sub_view != &vector_views.back() or options.is_enabled_lines()) {
+        if (out_sink->is_device_write_preferred(d_line_terminator.size())) {
+          out_sink->device_write(d_line_terminator.data(), d_line_terminator.size(), stream);
+        } else {
+          out_sink->host_write(line_terminator.data(), line_terminator.size());
+        }
+      }
     }
   } else {
     if (options.is_enabled_lines()) {

python/cudf/cudf/_lib/json.pyx

@@ -156,7 +156,7 @@ def write_json(
     bool include_nulls=True,
     bool lines=False,
     bool index=False,
-    int rows_per_chunk=8,
+    int rows_per_chunk=1024*256,  # 256K rows
 ):
     """
     Cython function to call into libcudf API, see `write_json`.

python/cudf/cudf/tests/test_json.py

@@ -187,14 +187,23 @@ def test_json_writer(tmpdir, pdf, gdf):
         assert_eq(pdf_string, gdf_string)
 
 
-def test_cudf_json_writer(pdf):
+@pytest.mark.parametrize(
+    "lines", [True, False], ids=["lines=True", "lines=False"]
+)
+def test_cudf_json_writer(pdf, lines):
     # removing datetime column because pandas doesn't support it
     for col_name in pdf.columns:
         if "datetime" in col_name:
             pdf.drop(col_name, axis=1, inplace=True)
     gdf = cudf.DataFrame.from_pandas(pdf)
-    pdf_string = pdf.to_json(orient="records", lines=True)
-    gdf_string = gdf.to_json(orient="records", lines=True, engine="cudf")
+    pdf_string = pdf.to_json(orient="records", lines=lines)
+    gdf_string = gdf.to_json(orient="records", lines=lines, engine="cudf")
+
+    assert_eq(pdf_string, gdf_string)
+
+    gdf_string = gdf.to_json(
+        orient="records", lines=lines, engine="cudf", rows_per_chunk=8
+    )
 
     assert_eq(pdf_string, gdf_string)